MSM6679A-110 Voice Recognition Processor MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Contents Description .......................................................................................................................................... Features ................................................................................................................................................ Functional and I/O Diagrams .......................................................................................................... Pin Descriptions ................................................................................................................................. Electrical Specifications ..................................................................................................................... Absolute Maximum Ratings ............................................................................................ Operating Conditions ....................................................................................................... DC Characteristics (VDD = 4.5 to 5.5 V, Ta = -40 to 80˚C) .......................................... AC Characteristics ............................................................................................................ Timing Diagrams .............................................................................................................. Functional Description ...................................................................................................................... Voice Recognition ............................................................................................................. SI Recognition ................................................................................................... SD Recognition ................................................................................................. Name Tag Recording ........................................................................................................ Audio Input Interface ....................................................................................................... Audio Output Interface .................................................................................................... Memory Interface .............................................................................................................. External Voice Synthesis Control ................................................................................... Serial Interface ................................................................................................................... MSM6679A-110 Slave-Mode API .................................................................................................... Command Summary ........................................................................................................ Command Descriptions ................................................................................................... Asynchronous Serial Protocol Example ........................................................................ 2 1 1 2 8 12 12 12 13 14 15 17 17 17 20 20 21 21 21 24 25 26 27 31 44 ¡ Semiconductor MSM6679A-110 VRP ¡ Semiconductor MSM6679A-110 Voice Recognition Processor SI/SD Voice Recognizer, Recorder/Player, and Speech Synthesizer DESCRIPTION The MSM6679A-110 Voice Recognition Processor (VRP) is a slave-mode device that performs five functions: speaker-independent (SI) voice recognition, speaker-dependent (SD) voice recognition, solid-state sound recording, sound playback, and speech synthesis. The highly integrated device also provides an on-chip memory controller, Flash memory interface, analog data conversion, Oki speech synthesizer interface, and pulse width modulation (PWM) sound output. For SI recognition, the MSM6679A-110 contains a vocabulary template in external memory. Pretrained SI vocabularies eliminate the need for laborious training, as usually required by SD products. The memory requirements are dependent on the size of the vocabulary. The MSM6679A110 can tolerate background noise, while providing high recognition accuracy. In its designated operating environment, the device achieves a typical recognition accuracy of >95% (using an Oki-defined test procedure). For SD recognition, the MSM6679A-110 stores SD vocabulary templates, as defined by the user, in external SRAM. The MSM6679A-110 can create SD vocabularies of up to 61 words each, with each word using approximately 50 bytes. In addition to providing voice recognition capabilities, the MSM6679A-110 integrates a solidstate recorder/player, speech synthesis functions, and a tone generator. ADPCM recording/ playback provides high quality sound and efficient memory utilization. The MSM6679A-110 can respond to spoken commands, verbally or with tones, via an on-chip speech synthesizer and tone generator. For larger speech-synthesis requirements, the MSM6679A-110 also provides a glueless MSM665x control interface for off-chip speech synthesis. The MSM6679A-110 can interface to any application or personal computer via a parallel or serial interface through an open, device-independent serial mode API (SMAPI). To accelerate code development, Oki supplies an evaluation kit, and assembly and C language programs for this product. FEATURES • SI recognition - Up to 20 - 25 words in each vocabulary - Multiple vocabulary support • SD recognition - Up to 61 words in each vocabulary - Multiple vocabulary support • Speech synthesis - Up to 2.3-sec internal and 27.6-sec external speech synthesis on-chip; sample looping and concatenation allows even longer phrases. - On-chip controller for MSM665x speech synthesizer - Standard beep tone outputs - Pulse code modualation (PCM) and adaptive differential pulse code modualation (ADPCM) voice or soundeffect output • Speech capture and playback - 28-kbps ADPCM speech compression • Serial ASCII command interface • 6944-Hz audio input sample rate for record andplayback • 10-kHz sample rate for voice recognition • 200-msec recognition latency • Flexible memory mapping for EPROM, FLASH, and SRAM • 32-MHz operation • Packages: 84-pin PLCC (QFJ84-P-S115) or 100-pin TQFP (TQFP100-P-1414-0.50-K) 1 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor FUNCTIONAL AND I/O DIAGRAMS Analog Recognition and System Serial Input Synthesis Engine Controller Interface PWM Vocabulary Algorithm Memory Memory Output External Speech Synthesis External Memory Control Control Figure 1. MSM6679A-110 Block Diagram ADC0 ~ ADC9 AD0 ~ AD7 A/D Interface VREF Serial-Mode MSM665x Interface PWM Output Serial Interface IC Reset and Oscillator Inputs NAR BUSY SI SD STROBE RESOUT VOICEOUT1 RXD1 TXD1 RES OSC0 OSC1 A8 ~ A15 ROMRD EA WRRAM RDRAM ALE ES LOADPGM ROMPAGE0 ROMPAGE1 RAMPAGE0 RAMPAGE1 A15FLIP Figure 2. MSM6679A-110 Logic Symbol 2 Memory Interface MSM6679A-110 Voice Recognition Processor N/C N/C N/C N/C N/C N/C RAMPAGE1 RAMPAGE0 LOADPGM N/C GND SD SI BUSY N/C ROMPAGE1 ROMPAGE0 STROBE A15FLIP N/C N/C ¡ Semiconductor 74 73 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58 57 56 55 54 , 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 N/C N/C N/C– A15 A14 A13 A12 A11 A10 A9 A8 VDD AD7 AD6 AD5 AD4 AD3 AD2 AD1 AD0 VOICEOUT1 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 75 76 77 78 79 80 81 82 83 84 1 2 3 4 5 6 7 8 9 10 11 VREF N/C+ N/C+ N/C N/C N/C N/C RXD1 TXD1 GND OSC0 OSC1 ALE ROMRD WRRAM RDRAM N/C– RESOUT ES NAR N/C N/C N/C N/C N/C N/C N/C+ RES EA VDD AVDD ADC0 ADC1 ADC2 ADC3 ADC4 ADC5 ADC6 ADC7 ADC8 ADC9 AGND Figure 3. MSM6679A-110 84-Pin PLCC Pinout 3 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor MSM6679A-110 Alphabetic Pin List Name 4 84-pin/ 100-pin Name 84-pin/ 100-pin Name 84-pin/ 100-pin Name 84-pin/ 100-pin Name 84-pin/ 100-pin Name 84-pin/ 100-pin AD0 34/28 A10 45/41 ADC3 4/92 BUSY SI 62/61 AD1 35/29 A11 46/42 ADC4 5/93 N/C+ 13,14,80/3,4,82 RAMPAGE1 68/68 STROBE 57/55 AD2 36/30 A12 47/43 ADC5 6/94 N/C– 28,51/20,47 RDRAM 27/19 TXD1 20/10 AD3 37/31 A13 48/44 ADC6 7/95 EA 81/83 VREF 12/2 AD4 38/32 A14 49/45 ADC7 8/96 ES AD5 39/33 A15 50/46 ADC8 9/97 GND AD6 40/34 A15FLIP 56/54 ADC9 10/98 LOADPGM AD7 41/35 ADC0 1/89 AGND 11/99 NAR 31/23 ROMRD 25/17 A8 43/39 ADC1 2/90 ALE 24/16 OSC0 22/13 RXD1 19/9 A9 44/40 ADC2 3/91 AVDD 84/87 OSC1 23/15 SD 63/62 61/60 RAMPAGE0 67/67 82/84 RES 30/22 RESOUT 29/21 VOICEOUT1 33/27 21,64/12,63 ROMPAGE0 58/56 WRRAM 26/18 66/66 ROMPAGE1 59/57 VDD 42,83/37,85 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Figure 4. MSM6679A-110 84-Pin Package Mechanical Drawing 5 76 N/C 77 N/C 78 N/C 79 N/C 80 N/C 81 N/C 82 N/C+ 83 RES 84 EA 85 VDD 86 N/C ¡ Semiconductor 87 AVDD 88 N/C 89 ADC0 90 ADC1 91 ADC2 92 ADC3 93 ADC4 94 ADC5 95 ADC6 96 ADC7 97 ADC8 98 ADC9 99 AGND 100 N/C MSM6679A-110 Voice Recognition Processor N/C 1 75 N/C VREF 2 74 N/C N/C+ 3 73 N/C N/C+ 4 72 N/C N/C 5 71 N/C N/C 6 70 N/C N/C 7 69 N/C N/C 8 68 RAMPAGE1 RXD1 9 67 RAMPAGE0 66 LOADPGM TXD1 10 N/C 11 65 N/C GND 12 64 N/C OSC0 13 63 GND 62 SD N/C 14 61 SI OSC1 15 60 BUSY ALE 16 ROMRD 17 59 N/C WRRAM 18 58 N/C RDRAM 19 57 ROMPAGE1 N/C– 20 56 ROMPAGE0 RESOUT 21 55 STROBE ES 22 54 A15FLIP Figure 5. MSM6679A-110 100-Pin TQFP Pinout 6 N/C 50 N/C 49 N/C 48 N/C– 47 A15 46 A14 45 A13 44 A12 43 A11 42 A10 41 A9 40 A8 39 N/C 38 VDD 37 N/C 36 AD7 35 AD6 34 AD5 33 AD4 32 AD3 31 AD2 30 AD1 29 51 N/C AD0 28 52 N/C N/C 25 N/C 26 53 N/C N/C 24 VOICEOUT1 27 NAR 23 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Figure 6. MSM6679A-110 100-Pin Package Mechanical Drawing 7 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor PIN DESCRIPTIONS Pin # 84-pin/ Pin Name Signal Type Description (not connected) Reserved. This pin is reserved for future use and should be left open. 100-pin -/1 N/C 1/89 ADC0 2/90 ADC1 3/91 ADC2 4/92 ADC3 5/93 ADC4 6/94 ADC5 7/95 ADC6 8/96 ADC7 9/97 ADC8 10/98 ADC9 Analog Input. These ten inputs are tied together and serve as the Analog input analog input. Signal conditioning, via a bandpass filter and gain circuit, is required before this input. Analog Ground. This pin provides an analog ground point, allowing 11/99 AGND Analog ground independent grounding of the analog and digital circuitry. Separate grounds reduce the impact of digital switching noise on analog sampling accuracy. Analog Reference Voltage. The MSM6679A-110's on-chip A/D 12/2 VREF Reference voltage converter uses this analog reference voltage when converting an analog signal into digital samples 13/3 N/C+ 14/4 N/C+ 15/5 N/C 16/6 N/C 17/7 N/C Input Reserved. These pins are reserved for future use and must be tied to VDD. (do not connected) Reserved. These pins are reserved for future use and must be left open. 18/8 N/C 19/9 RXD1 20/10 TXD1 Output -/11 N/C (not connected) 21/12 GND Ground Input Serial Port Receive. This is the receive data line for serial port. Serial Port Transmit. This is the transmit data line for serial port. Reserved. This pin is reserved for future use and should be left open. Ground. Oscillator 0/External Clock. When the MSM6679A-110 uses a crystal 22/13 OSC0 Input oscillator, this input is the oscillator input pin. The pin is then connected to one side of a crystal and load capacitor. When used with an external clock, the external clock is applied to this input. -/14 N/C (not connected) Reserved. This pin is reserved for future use and should be left open. Oscillator 1. When the MSM6679A-110 uses a crystal oscillator, this 23/15 OSC1 Output output is the oscillator output pin. The pin is then connected to one side of a crystal and load capacitor. When used with an external clock, this output is left unconnected. 24/16 8 ALE Output Memory Address Latch Enable. An external memory latch is controlled by this signal, the address latch enable output. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Pin # 84-pin/ Pin Name Signal Type Description 100-pin ROM Read. This is a strobe signal for direct connection to an external 25/17 ROMRD Output ROM's READ input. When asserted LOW, this signal indicates that the MSM6679A-110 is ready to read data from the ROM. RAM Write. This is a strobe signal for direct connection to an external 26/18 WRRAM Output RAM's WR input. When asserted LOW, this signal indicates that the MSM6679A-110 is ready to write data to RAM. RAM Read. This is a strobe signal for direct connection to an external 27/19 RDRAM Output 28/20 N/C– Input 29/21 RESOUT Output 30/22 ES Output RAM's RD input. When asserted LOW, this signal indicates that the MSM6679A-110 is ready to read data from RAM. Reserved. This pin is reserved for future use and must be tied to GND. MSM665x Reset. This pin provides a reset signal for an external speech synthesis engine. Flash Bank Control (Extended Segments). This is the control signal for flash memory banking. MSM665x Next Address Request. This pin signals to the 31/23 NAR Input MSM6679A-110 that the external speech synthesis engine is ready for another command. 32/24 N/C (do not connect) Reserved. These pins are reserved for future use and must be left open. -/25,26 N/C (not connected) Reserved. These pins are reserved for future use and should be left open. 33/27 VOICEOUT1 Output 34/28 AD0 35/29 AD1 36/30 AD2 37/31 AD3 38/32 AD4 39/33 AD5 40/34 AD6 41/35 AD7 Voice Out. This pin is the PWM output for speech synthesis, voice sample playback, and voice prompts. An external integrator must be used to convert this to an analog signal. Memory Address/Data Bus. These are multiplexed address/data lines Bidirectional I/O address bits are not multiplexed). -/36 N/C (not connected) 42/37 VDD Digital Power -/38 N/C (not connected) 43/39 A8 44/40 A9 45/41 A10 46/42 A11 47/43 A12 48/44 A13 for the eight data bits and the lower eight address bits (the upper eight Outputs Reserved. This pin is reserved for future use and should be left open. Power. Reserved. This pin is reserved for future use and should be left open. Memory Address Bus. These are the upper eight address pins. 9 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Pin # 84-pin/ Pin Name Signal Type Description 100-pin 49/45 A14 50/46 A15 51/47 N/C– 52/48 N/C 53/49 N/C -/50,51 N/C 54/52 N/C 55/53 N/C 56/54 A15FLIP Output 57/55 STROBE Output 58/56 ROMPAGE0 59/57 ROMPAGE1 60/58 N/C (do not connect) Reserved. This pin is reserved for future use and must be left open. -/59 N/C (not connected) Reserved. This pin is reserved for future use and should be left open. Outputs Memory Address Bus. These are the upper eight address pins. Input Reserved. This pin is reserved for future use and must be tied to GND. (do not connect) Reserved. These pins are reserved for future use and must be left open. (not connected) Reserved. These pins are reserved for future use and should be left open. (do not connect) Reserved. These pins are reserved for future use and must be left open. Outputs Memory Address A15 Flip. This signal inverts the A15 address signal for 32-Kbyte bank switching on the local memory bus. MSM665x Strobe. This output provides the LOAD signal for an external speech synthesizer. ROM Page Select. These signals select one of four 64-Kbyte ROM pages. MSM665x Busy. When using an external MSM665x device, this pin 61/60 BUSY Input monitors the MSM665x BUSY signal and connects directly to the MSM665x BUSY signal output. MSM665x Serial Clock. This MSM6679A-110 output connects to the 62/61 SI Output 63/62 SD Output 64/63 GND Digital Ground Ground. MSM665x SI input. The SI pin is the MSM665x serial clock input pin. MSM665x Serial Data. This MSM6679A-110 output connects to the MSM665x SD input. The SD pin is the MSM665x serial data input pin. -/64 N/C (not connected) Reserved. This pin is reserved for future use and should be left open. 65/65 N/C (do not connect) Reserved. This pin is reserved for future use and must be left open. 66/66 LOADPGM Output 67/67 RAMPAGE0 68/68 RAMPAGE1 69/69 N/C 70/70 N/C 71/71 N/C 72/72 N/C 73/73 N/C 74/74 N/C -/75,76 N/C Load Program. This signal allows the MSM6679A-110 to write data to program memory. When asserted low, this signal should set the program memory in write mode. 10 Output RAM Page Select. These signals support selection of one out of four RAM pages. Each page is 64kbytes in size. (do not connect) Reserved. These pins are reserved for future use and must be left open. (not connected) Reserved. These pins are reserved for future use and should be left open. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Pin # 84-pin/ Pin Name Signal Type Description (do not connect) Reserved. These pins are reserved for future use and must be left open. Reserved. This pin is reserved for future use and must be tied to VDD. 100-pin 75/77 N/C 76/78 N/C 77/79 N/C 78/80 N/C 79/81 N/C 80/82 N/C+ Input 81/83 RES Input 82/84 EA Input 83/85 VDD -/86 N/C (not connected) 84/87 AVDD Analog power supply -/88 N/C (not connected) Reserved. This pin is reserved for future use and should be left open. -/100 N/C (not connected) Reserved. This pin is reserved for future use and should be left open. MSM6679A-110 Reset. External logic should assert this power-on reset signal LOW when power is applied to the MSM6679A-110. External ROM Address Select. This control signal enables external ROM execution. This signal is usually connected to ROMPAGE1 and a pullup resistor. Positive digital supply Power. Reserved. This pin is reserved for future use and should be left open. Analog Power. 11 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor ELECTRICAL SPECIFICATIONS Absolute Maximum Ratings Parameter Digital power supply voltage Symbol Conditions Value –0.3 to +7.0 VDD Input voltage VI –0.3 to VDD +0.3 Output voltage VO –0.3 to VDD +0.3 Analog power voltage AVDD Analog reference voltage VREF Analog input voltage VAI Power dissipation Storage temperature 1. PD TSTG Unit GND = AGND = 0 V –0.3 to VDD +0.3 V –0.3 to AVDD +0.3 –0.3 to VREF Ta = 85˚C, per package 1300 max. Ta = 85˚C, per pin 50 max. — –50 to +150˚C mW ˚C Permanent device damage may occur if ABSOLUTE MAXIMUM RATINGS are exceeded. Functional operation should be restricted to the conditions as detailed elsewhere in this data sheet. Exposure to absolute maximum rating conditions for extended periods may affect device reliability. Operating Conditions Symbol Conditions Value Digital power supply voltage VDD fOSC = 32 MHz 4.5 to 5.5 Analog power supply voltage AVDD VDD = AVDD Analog reference voltage VREF AVDD –0.3 to AVDD Analog input voltage VAI AGND to VREF Parameter Unit 4.5 to 5.5 V Storage holding voltage VDDH fOSC = 0 MHz 2.0 to 5.5 Operating frequency fOSC VDD = 5 V ±10% 32 MHz Ambient temperature Ta ˚C Fan-out 12 N — –40 to 85˚C MOS load 20 TTL load, AD0 ~ AD7 2 TTL Load, all other outputs 1 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor DC Characteristics (VDD = 4.5 to 5.5 V, Ta = -40 to 85˚C) Parameter Symbol Condition Low-level input voltage VIH VIL Typ [1] Max 2.2 — VDD +0.3 Applied to OSC0 0.85 × VDD — VDD +0.3 Applied to all other I/O 0.80 × VDD — VDD +0.3 Applied to AD0-AD7 –0.3 — 0.8 Applied to OSC0 –0.3 — 0.15 × VDD Applied to all other I/O –0.3 — 0.2 × VDD VDD –0.4 — — VDD –0.4 — — — — 0.4 — — 0.4 — — 1/–1 VI = VDD/0 V, applied to RES — — 1/–250 VI = VDD/0 V, applied to OSC0 — — 15/–15 VO = 2.4 V, applied to AD0-AD7 –2 — — VO = 2.4 V, applied to all other I/O –1 — — VO = 2.4 V, applied to AD0-AD7 10 — — VO = 2.4 V, applied to all other I/O 5 — — — ±2 — 5 — — 7 — During voice input — — 4 mA When voice input is halted — — 10 µA fOSC = 32 MHz, no load — 55 75 mA Output current = 400 mA, applied High-level output voltage VOH to AD0-AD7, ALE, and ROMRD Output current = 200 mA, for all other I/O Output current = 3.2 mA, applied Low-level output voltage VOL to AD0-AD7, ALE, and ROMRD Output current = 1.6 mA, for all other I/O VI = VDD/0 V, applied to Ain, EA, Input leak current IIH, IIL Input current High-level output current IOH Low-level output current IOL Output leakage current ILO Input capacitance CI Output capacitance CO Analog reference power supply voltage Power consumption 1. IREF IDD Unit Min Applied to AD0-AD7 High-level input voltage Rated Value FLOAT, and RESTART VO = VDD/0 V f = 1 MHz, Ta = 25˚C V µA mA µA pF Typical condition is 5 V 25˚C. 13 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor AC Characteristics External Program Memory Control (VDD = 4.5 to 5.5 V, Ta = -40 to 85˚C) Symbol Condition Min. Max. Clock pulse width (OSC) Parameter tOW — 15.625 — ALE pulse width tAW 36.875 — ROMRD pulse width tPW 177.5 — ROMRD pulse delay time tPAD 10.625 20.625 Low address set-up time tAAS 21.25 41.25 Low address hold time tAAH 10.625 20.625 High address delay time tAAD 15.625 25.625 High address hold time tAPH 15.625 25.625 Instruction set-up time tIS 35 — Instruction hold time tIH 0 25.625 CL = 50 pF Unit ns External Data Memory Control (VDD = 4.5 ~ 5.5 V, Ta = -40 ~ 85˚C) Parameter Symbol Condition Min. Max. Clock pulse width (OSC) tOW — 15.625 — ALE pulse width tAW 36.875 — RDRAM pulse width tRW 177.5 — WRRAM pulse width tWW 177.5 — RDRAM pulse delay time tRAD 10.625 20.625 WRRAM pulse delay time tWAD 10.625 20.625 Low address set-up time tAAS 21.25 41.25 Low address hold time tAAH 10.625 20.625 High address set-up time tAAD 15.625 25.625 15.625 25.625 High address hold time tARH, tAWH CL = 50 pF Memory data set-up time tMS 35 — Memory data hold time tMH 0 5.625 Data set-up time tDD 15.625 25.625 Data hold time tDH 15.625 25.625 14 Unit ns ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Timing Diagrams tOW tOW CLK tAW ALE tPAD tPW ROMRD tAAS AD0 - AD7 tAAH tIS PC0 - 7 tIH INST0 - 7 tAAD tAPH PC8 - 15 A8 - A15 Figure 7. ROM Read Timing tOW tOW CLK tAW ALE tRAD tRW RDRAM tAAS AD0 - AD7 tAAH tMS RAP0 - 7 INST0 - 7 tAAD A8 - A15 tMH tARH RAP8 - 15 Figure 8. RAM Read Timing 15 MSM6679A-110 Voice Recognition Processor tOW ¡ Semiconductor tOW CLK tAW ALE tWAD tWW WRRAM tDD tAAS AD0 - AD7 tDH tAAH RAP0 - 7 DOUT0 - 7 tAAD A8 - A15 tAWH RAP8 - 15 Figure 9. RAM Write Timing 16 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor FUNCTIONAL DESCRIPTION Voice Recognition The MSM6679A-110 performs both SD and SD recognition. SI vocabularies are embedded in the MSM6679A-110. For SD recognition, each recognized phrase must be enrolled in the MSM6679A110’s vocabulary by creating a composite template from multiple recordings of the same phrase. Then the composite tempalte is stored in SRAM or FLASH memory. During both SI and SD recognition, the MSM6679A-110 performs the following steps: 1. After external band-pass filtering, the MSM6679A-110 converts the analog signal to PCM samples. 2. The MSM6679A-110 extracts significant features from the sample data by frequency and time-domain analysis. 3. The MSM6679A-110 compares the analyzed input with the reference data for each signal, weighing the significance of similarities according to control software parameters. A score (expressed as distance) is generated for each phrase. 4. he vocabulary phrase that achieves the highest score (or lowest distance) is judged to match the input phrase, assuming that the score exceeds a predetermined threshold. 5. Via a special command, the MSM6679A-110 can also return the scores of the input against all defined vocabulary phrases for SI or SD recognition. This feature allows external host software to select the next best match, if the closest match is not contextually logical. SI Recognition Oki supplies the MSM6679A-110 with predefined SI vocabularies which Oki builds from hundreds of utterances by a wide variety of speakers. SI vocabularies are limited to 25 words or less, which allows the MSM6679A-110 to achieve a net accuracy of >95%, even in noisy conditions. SI vocabularies are grouped into sub-vocabularies of ≤15 words, to maintain the highest accuracy. Similar words in any one sub-vocabulary can cause substitution errors. Oki Semiconductor’s standard cellular vocabulary is intended for an automotive environment with a far-talk microphone. This vocabulary may work adequately in other conditions, such as an office or outside, but recognition performance may be degraded. MSM6679A-110 Cellular SI Recognition Vocabulary Sub-Vocabulary 1 Sub-Vocabulary 2 Sub-Vocabulary 3 Phrase Index Phrase Index Phrase Index Phrase Index Store 1 One 1 Eight 8 Yes 1 Dial 2 Two 2 Nine 9 No 2 Delete 3 Three 3 Zero Ah Cancel 3 Directory 4 Four 4 Oh Bh — — — — Five 5 Stop Ch — — — — Six 6 Clear Dh — — — — Seven 7 — — — — 17 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor MSM6679A-110 Control Vocabulary Sub-Vocabulary 1 Sub-Vocabulary 2 Phrase Index Phrase Index A/C 1 Low 1 Fan 2 Medium 2 Temperature 3 High 3 Timer 4 Increase 4 Service 5 Decresse 5 Help 6 Set 6 Select 7 Reset 7 — — Cancel 8 — — Clear 9 — — Recall A — — On B — — Help C MSM6679A-110 Direction Vocabulary Sub-Vocabulary 1 Phrase Index Up 1 Down 2 Left 3 Right 4 Formard 5 Reverse 6 Faster 7 Slower 8 Start 9 Stop A Cancel B MSM6679A-110 Browse Vocabulary Sub-Vocabulary 1 18 Sub-Vocabulary 2 Phrase Index Phrase Index Phrase Index Phrase Index Phrase Index Up 1 Next 5 Home 9 Set 1 On 5 Down 2 Previous 6 — — Reset 2 Play 6 Left 3 Select 7 — — Start 3 Lock 7 Right 4 Cancel 8 — — Stop 4 Cancel 8 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor MSM6679A-110 Japanese Navigation Vocabulary Sub-Vocabulary 1 Sub-Vocabulary 2 Sub-Vocabulary 3 Sub-Vocabulary 4 Phrase Index Phrase Index Phrase Index Phrase Index Genzaichi 1 Ue 1 Hyoujun 1 Hai 1 Jiaku 2 Shita 2 Kakudai 2 Iie 2 Kaisya 3 Hidari 3 Shukushou 3 Ofu 3 Houi 4 Migi 4 Zentai 4 — — Sentaku 5 — — Kaiten 5 — — Yuudou 6 — — Kyori 6 — — Nabi 7 — — Hosei 7 — — — — — — Teisei 8 — — MSM6679A-110 Japanese Celluar Vocabulary Sub-Vocabulary 1 Sub-Vocabulary 2 Phrase Index Phrase Index Phrase Index On 1 Ichi 1 Kyuu 9 Ofu 2 Ni 2 Zero A Daiyaru 3 San 3 Sharp B Tansyuku 4 Yon 4 Star C Denwacho 5 Go 5 Kakunin D Kakunin 6 Roku 6 Touroku E Nabi 7 Nana 7 Rei F — — Hachi 8 — — SI vocabulary generation starts with collecting reference utterances from ≥400 speakers with: • An equal mixture of males and females • Accents from all regions of the country of intended use • ~15% non-native speakers. The samples should be generated from a randomly-ordered list, with each word spoken twice and with a dummy word at the beginning and end. There must be >2 sec between each sample for accurate data processing. To provide the audio fidelity required for high-quality recognition training, a DAT recorder, together with the microphone that will be used in the final application, is required. To ensure data integrity, data is submitted to Oki after collecting samples from the first 20 speakers for initial screening. If acceptable, then the remaining collection may proceed. If substitution errors are possible, collection of spare words during initial collection is recommended. For example, alternate words to “Stop” and “Top” could be “Halt” and “First.” Collections should contain a wide variety of the background sound conditions that will exist during actual usage. For example, if the collection is for use in an automobile, conditions such as vehicle speed, road conditions, various window opening positions, heater or AC blower speeds and radio volumes should be varied during the collection. The signal-to-noise ratio should be maintained at ≥ 20dB. 19 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor To achieve high accuracy rates, phrase selection, data collection, background initialization strategy, and control software need careful consideration. There are no published standards for recognition accuracy. Oki defines accuracy by: Accuracy = 100% - ERATE ERATE = ESUB + 1/2 EREJ with the following definitions: Parameters for Recognition Accuracy Name Substitution Error Symbol Condition ESUB Most critical type error, e.g., Say "Five", recogrize "Nine" Rejection Error EREJ Word not recognized, opportunity for operator to repeat Gap Error EGAP Word spoken before recognizer ready Time-Out Error ETME Spurious Response Error ESPU Word length is too long Sourd or imvalid word classfied as a valid word (i.e., drop handset or speak wong word) A typical target accuracy of 97% is achieved with a 3% ERATE , composed of a 1.5% ESUB rate and a 3%EREJ rate. SD Recognition In SD recognition mode, the MSM6679A-110 can be trained to recognize up to 61 words. The MSM6679A-110 can support multiple speakers by switching vocabularies, but only one speaker’s vocabulary should be active at one time. The end user enrolls a phrase in the MSM6679A-110’s vocabulary by recording the phrase three times or more. The host Micro Controller Unit (MCU) controls the number of times each phrase in enrolled. Generally, higher recognition accuracy is achieved with each additional enrollment. The word set is made more robust by pronouncing each phrase slightly differently during initial enrollment. In addition to enrollment training, adaptive template updating can drive the accuracy towards 100%. The host MCU updates templates by first asking the speaker to confirm a recognized phrase with a “yes” or “no” response, and subsequently updating the template for corresponding words. The use of name tags (see next paragraph) facilitates this process. Name Tag Recording To facilitate SD recognition, the MSM6679A-110 supports recording and playback of name tags. Name tags are used to confirm correct responses in SD recognition. For example, in a phone dialer application, the user associates a “name” (which is recorded into memory) with a phone number. The MSM6679A-110 then plays back the name tag so that the user can verify that the recognized phrase is the correct one. The VRP stores names tags in memory using an ADPCM compression algorithm with 28 kbps of speech. The length of a name tag is controlled with a command from the users host MCU program. The maximum number of name tags possible is 61, but the actual number is dependent upon record time and memory available. See the section on memory interface for more detail. 20 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Audio Input Interface A critical item for high-accuracy speech recognition is correct design of the audio input circuit. A circuit with appropriate gain and frequency responses must be placed between the microphone and MSM6679A-110’s A/D input. Oki recommends input gain and a band pass filter with the following characteristics: • Four pole Chebyshev high-pass filter, 3 dB point at 225 Hz • Dual-pole low-pass filter, 3 dB point at 4250 Hz • Midband gain of 46 dB at 1000 Hz The above gain and filter characteristics are obtained by using a rail-to-rail quad CMOS op-amp and one-half supply rail splitter to bias the input signal at 2.5 V nominal. The MSM6679A-110 uses multiple analog inputs to improve sampling quality. An on-chip analogy to digital (A/D) conversion unit transforms the analog signal to a digital data stream. Audio Output Interface The MSM6679A-110 also provides the VOICEOUT1 PWM output. The MSM6679A-110 uses ADPCM to generate voice or sound-effect output. ADPCM represents an improvement over conventional PCM techniques in that it adaptively changes the quantizer step (scale factor) to suit the waveform being encoded. The result is more efficient memory usage with no loss of quality. Careful selection of the components for internal and external output filters and amplifiers is recommended. An incorrect choice would impair the original quality. This consideration equally includes: • Careful separation of analog and digital lines • Grounding of analog lines at both ends • Further adequate separation from high-speed digital circuits to avoid distortions thereof Memory Interface The memory control section manages RAM and/or ROM devices in two 64-Kbyte memory spaces, in conjunction with internal memory for voice templates and working memory. Some versions work with no external memory, some have some external RAM, some use only external EPROM, and some use external memory in conjunction with both internal ROM and RAM. The MSM6679A-110 requires a minimum of 32 Kbytes SRAM and 16 Kbytes ROM. The following table shows vocabulary sizes and playback facilities for various configurations. 21 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Typical Configurations Recognition MSM6679A-110 Application Controller Vocabulary Sound Playback (Words) (sec) [1] MSM665x MSM6679A-110 MSM6679A-110 Playback Speech Speech Interface Record Playback SI SD 25 61[2] 2.3 9.2 OK — OK 50 61[2] 2.3 — OK — OK Internal External Memory Size (bytes) EPROM Flash SRAM 64K — 32K — 128K 32K 25 61 2.3 27.6 OK OK OK Telephone 50 61 2.3 18.4 OK OK OK Dialer 75 61 2.3 — OK OK OK 100 61 2.3 — OK OK OK 61[3] 61 2.3 36.8 OK — OK — — 64-384K 12 61[2] 1.15 OK — — 16K — 32K Computer Peripheral Minimum Configuration 1. 2. 3. Phrase chaining features usually permit much longer overall playback durations; not including external speech synthesizer. SD recognition vocabularies are volatile in these configurations. Per download. Vocabulary swapping by host permits unlimited vocabulary size. The MSM6679A-110 supports up to 64 Kbytes of RAM per bank, and up to 64 Kbytes of ROM per bank in separate memory spaces. The 8-bit data bus is multiplexed with the lower eight address bits; the upper eight address bits are not multiplexed. To demultiplex the address and data bits during all read and write cycles, the MSM6679A-110 requires an external octal latch, such as the 74H373. The MSM6679A-110’s Address Latch Enable (ALE) signal controls the octal latch. For accessing the ROM and RAM address spaces, the MSM6679A-110 provides the separate Write RAM (WRRAM), Read RAM (RDRAM), and ROM Read (ROMRD) signals. The RDRAM and ROMRD signals connect directly to Output Enable (OE) control signal inputs on the RAM and ROM, respectively. The WRRAM signal connects directly to the Write Enable (WE) control signal input on the RAM. The following diagrams show the memory maps for the MSM6679A-110. In all MSM6679A-110 memory maps, the DL data memory space must be in RAM. The DH data memory space and PH program memory space can either be implemented in ROM, EPROM, FLASH, RAM, or PROM. In standalone applications, flash memory can be used for recording and subsequent playback of voice prompts (e.g., the user’s name) and user sounds (e.g., DTMF dial tones, etc.). Figure 10 shows the configuration for writing to flash memory used when writing SD templates or when flash is used for data memory. 22 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Program Memory 00000 MSM6679 Internal Program Memory BFFF C000 FFFF PHH FLASH 00000 03FFF 04000 07FFF 08000 08FFF 0C000 0FFFF 10000 13FFF 14000 17FFF 18000 18FFF 1C000 1FFFF F0 SRAM Data Memory 00000 00000 DL F1 F2 07FFF 08000 SO 07FFF 08000 DH F3 0FFFF S1 0FFFF 10000 F4 S2 F5 Hidden 17FFF 18000 F6 S3 F7 1FFFF Figure 10. MSM6679A-110 Program/Data Memory Map (LOADPGM = “0”) Figure 11 shows the memory map during all other modes of operation. Program Memory 00000 MSM6679 Internal Program Memory BFFF C000 FFFF PHH FLASH 00000 03FFF 04000 07FFF 08000 08FFF 0C000 0FFFF 10000 13FFF 14000 17FFF 18000 18FFF 1C000 1FFFF F0 DL F1 F2 07FFF 08000 DH F3 SRAM Data Memory 00000 0FFFF 00000 07FFF 08000 0FFFF 10000 17FFF 18000 1FFFF S0 S1 S2 S3 F4 F5 F6 F7 Figure 11. MSM6679A-110 Program/Data Memory Map (LOADPGM = “1”) Figure 12 shows the details of the external memory allocation of the MSM6679A-110. 23 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor FLASH FLASH 00000 Reserved 04AD0 Default Working SD Templates 05480 Working Name Tag Pointer Table 05700 Alternate SD Templates 08000 Down load RAM Bank 0F300 (F510*) Alternate SD Templates 10000 SI First (F509*) 07300 SD First 07D80 NTP First Name Tag Block Address 08000 000 10000 100 Name Tag Data 18000 SI Last (F501*) 200 1F900 SD Last 1FD80 NTP Last 1FFFF 2F6 Reserved 18000 Buffer RAM Bank (F520*) 2FB 1F300 2FF 1FFFF Reserved *Denotes commands to select blocks Figure 12. MSM6679A-110 External Memory Map External Voice Synthesis Control The MSM6679A-110 is capable of interfacing to the MSM665x family of Oki ROM, OTP, or external EPROM speech synthesizers, allowing for up to 260 seconds of high-quality voice and sound effects. The following table indicates the speech capabilities of the MSM665x family. MSM665x Family Characteristics Type Maximum Speech Duration[2] Data ROM Capacity[1] fSAM = 4.0 kHz fSAM = 6.4 kHz fSAM = 8.0 kHz fSAM = 16.0 kHz fSAM = 32.0 kHz MSM6650 64 Mbits[3] >1 hour >40 minutes MSM6652 288 Kbit 16.9 sec MSM6653 544 Kbit 31.2 sec MSM66P54[4] 1 Mbit MSM6654 1 Mbit >30 minutes >15 minutes >8 minutes 10.5 sec 8.4 sec 4.2 sec 2.1 sec 19.5 sec 15.6 sec 7.8 sec 3.9 sec 63.8 sec 39.9 sec 31.9 sec 15.9 sec 7.9 sec 63.8 sec 39.9 sec 31.9 sec 15.9 sec 7.9 sec MSM6655 1.5 Mbit 96.5 sec 60.3 sec 48.2 sec 24.1 sec 12.0 sec MSM66P56[5] 2 Mbit 129.1 sec 80.7 sec 64.5 sec 32.2 sec 16.1 sec MSM6656 2 Mbit 129.1 sec 80.7 sec 64.5 sec 32.2 sec 16.1 sec MSM6658 4 Mbit 258 sec 161.4 sec 129.1 sec 64.5 sec 32.2 sec 1. 24 Actual ROM area in MSM6652, MSM6653, MSM6654, MSM6655, and MSM6656, MSM6658, MSM66P54, MSM66P56 is smaller by 22 Kbits. ¡ Semiconductor 2. 3. 4. 5. MSM6679A-110 Voice Recognition Processor Longer speech patterns can be created by chaining and repeating existing speech samples. Via external ROM only (no on-chip ROM available). One-Time-Programmable (OTP) version of MSM6654. See the MSM66P54 data sheet for more information. One-Time-Programmable (OTP) version of MSM6656. See the MSM66P56 data sheet for more information. The MSM665x interface consists of the following signals: • BUSY - Asserted LOW during MSM665x device playback. The MSM6679A-110 F50Bh and F10100xxh commands select this signal for MSM665x command polling. • NAR - Next Address Request status signal. By default, the MSM6679A-110 uses this signal to poll commands to the MSM665x. The F51Bh, F480h, and F440h commands select NAR for polling. • SI - Serial Input Clock. • SD - Serial Data Out. • STROBE - Initiates speech synthesis. • RESOUT - Initializes device when asserted LOW. The MSM6679A-110 F480h command generates this signal. Serial Interface The MSM6679A-110 supplies a serial interface suitable for connection to an RS-232C serial port buffer or equivalent. The serial interface uses one MSM6679A-110 input (RXD) and one MSM6679A-110 output (TXD). The interface operates at 9600 Baud with: • 8 data bits • 1 start bit • 1 stop bit • No parity • No handshake A host processor sends serial ASCII commands to the MSM6679A-110 and receives serial ASCII responses based on voice input responses. 25 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor MSM6679A-110 SLAVE-MODE API This section describes the slave-mode Applications Protocol Interface (API) between a host MCU and the MSM6679A-110. The slave-mode API offers the following features: • Direct slave-mode control voice recognition, sound recording and playback, and sound synthesis • Serial port interfaces • Simple procedures for downloading and uploading data • ASCII format • Comprehensive return codes and error reporting The host MCU selects the active speech recognition vocabulary, speech responses, and controls all actions required to implement an interactive voice response system. The MSM6679A-110 performs speech recognition, based on the vocabulary selected by the host, and returns digital codes representing the most probable match of the current utterance to an individual utterance in the selected vocabulary. The MSM6679A-110 can also respond with “name tags.” Name tags can be fixed words, phrases or sound effects, or can be words, phrases or sound effects that have been interactively recorded by the user. The API supports both serial and parallel interfaces. The MSM6679A-110 returns each response using the same interface through which the most recent message was received. The user can thus connect and use both interfaces. For all messages, the parallel interface uses 8-bit binary values, while the serial interface represents each 8-bit value with two hexadecimal digits coded in ASCII. When downloading and uploading data, the MSM6679A-110 uses a stream of 8-bit binary values on both parallel and serial interfaces. The serial-mode interface uses a 9600-baud UART with 1 start bit, 8 data bits, and 1 stop bit. There is no parity or handshaking. Serial-interface messages are of variable length, but consist of an even number of bytes. The serial interface echoes all received ASCII characters immediately back to the host MCU. Messages are of variable length. All messages consist of an even number of bytes. Opcodes consist of exactly four bytes, with values between F000h and FEFEh. Operand bytes may take values from 0000h to FFFFh. The MSM6679A-110 issues a return code for many of the host commands. The return code generally consists of the same opcode, followed by data indicating success of failure of the operation. Opcodes are organized into the following categories: • Purge • Set parameter • Initialize • Recognize • Speak • Request • Record • SD recognition control The following tables summarize available opcodes and provide detailed descriptions of the opcode functions. 26 ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Command Summary Function Opcode (Hex) Description Default (Hex) Purge F000 Clear MSM6679A-110 input stack — Set parameter F102 xxxx F103 xxxx F104 xxxx F11x F12x F130 xxxx F440 Set SP/SI origin to xxxx. Set SD origin. Set triggering origin. Set IRQ level to IRQ x. Set SD SP table to table x. Select triggering table. Set ISA mode. 8000 4A00 F100 0005 F123 0101, 0202... Disabled. F2xx mod 80 F2xx mod 40 F2xx mod 20 F2xx mod 10 F2xx mod 8 F2xx mod 4 F2xx mod 2 F2xx mod 1 Initialize background estimation. Wait for F3h command after each response. Beep after each triggered utterance Reserved Set speech response level to default. Send acknowledge after each speech output response. Only detect triggers. Initialize SD parameter table and name tags. Disabled. Disabled. Disabled. Disabled. Enabled. Enabled. Disabled. Load from first FLASH. F300 F301 to F33F F340 F341 Stop listening (recognition). Start SI recognition. Start SD recognition. Sort SD recognition distances, return index to utterance with least distance. Update SD enrollment. Request recognition parameter upload to host. Sort SD recognition distances, return index and distance to utterance with least distance Sort SD recognition distances, return all distances. Sort SD recognition distances, return minimum and maximum energy values. Sort SD recognition distances, return all energy values and distances. — — — — Play back name tag from external memory. Play back sound from internal memory. Play 50-ms beep. Pause for 0.2 sec. Initialize MSM665x IC, set MSM665x busy mode OFF, select FLASH SI recognition. Play back one of 127 phrases in external MSM665x device. Set MSM665x busy mode ON. Set 6654 NAR mode Set output volume (03h = minimum, FEh = maximum). — — — — — Status request. Select last FLASH bank for SI recognition. Select download RAM bank for speaker independent/signal processing (SI/SP) template area. Select buffer RAM bank for SI/SP. Copy download RAM bank to buffer RAM bank Save download RAM bank templates in first FLASH. (8000 - F2FF) — F509 F509 Initialize Recognize F342 F343 F344 F351 F361 F371 Speak F401 to F43D F441 to F47C F47E F47F F480 F481 - F4FF F50B F51B FE03 to FEFE F500 F501 F510 Request F520 F522 F513 — — — — — — — OFF ON FE80h — — — 27 MSM6679A-110 Voice Recognition Processor Function Opcode (Hex) F514 SD Recognition Control Default (Hex) — 0051 0000 01FF — — — F50D F51D F50E F50F FA01 ~ FA3D Set name tag length, set MSM665x busy mode ON. Set name tag record origin Set name tag record end Clear name tag table in SRAM (5480 - 56FF). Recall last saved name tag table. Recall name tag pointers from last FLASH (FD80-FFFF→5480-56FF) Save name tag table from SRAM to FLASH. Save name tag pointers in last FLASH (5480-56FF→FD80-FFFF) Set record volume high. Set record volume normal (default). Record name tag 01h - 3Dh. F6xx F9xx FB00 FC00 F521 Set SD pointer to segment xxh. Search for SD utterance xxh. Enroll SD utterance selected by search command (F9xx). Erase utterance from SD vocabulary. Clear SDR table (4A00 - 547B) — — — — — F516 Record Description Get download RAM bank templates from the first FLASH (8000 - FFFF) Save download RAM bank templates is last FLASH (8000 - F2FF) Get download RAM bank templates from last FLASH (8000 - FFFF) Download/upload. Select/jump. Retrieve MSM6679A-110 firmware revision. Initialize background (BG) noise level. Retrieve vocabulary and trigger table revision number. Save SD templates from download RAM to first FLASH. Save SDR templates in last FLASH. (4A00-547B→F300-FD7F) Recall SD templates from first FLASH to download RAM. Get SDR Templates from last FLASH (F300-FD7B→4A00-547B) Select first FLASH bank for SI recognition. F515 Request ¡ Semiconductor F502.... F503 xxxx F504 F505 F506 F507 F517 F508 F518 F509 F101 00xx F105 F106 F50A F50C F51C — — — — 3136 — 3330 — — — — F509 — — F50F F50F — Response Summary Command Result after Parameter Set Operands F101h 00 tm F102h AdH AdL F103h AdH AdL F104h AdH AdL F11Xh F12Xh Description Record time = tm*14 msec. High and low bytes of SP/SI origin address. High and low bytes of SD origin address. High and low bytes of triggering origin address. IRQ Xh selected. SP table Xh selected. F280h F240h F220h F210h Initialization Acknowledgment F208h F204h F202h F201h Invalid message received. Sample data over-run. [1] 32-Kbyte block boundary violation error. Unclassified download/upload error. Divide-by-zero error. Select/jump error. Invalid SP header or table. Reserved. Speech Ack Speech acknowledgment. [2] 28 F400h ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Response Summary (Continued) Command Status [3] SI Recognition Result [5] Operands Description F500h F501h F520h F540h F560h F580h F5A0h F5C0h F5E0h F5F0h MSM6679A-110 ready. Operation complete. Operations complete; MSM6679A-110 disabled (vocabulary 0). MSM6679A-110 waiting for start command. MSM6679A-110 waiting for end trigger. MSM6679A-110 processing recognition. Download/upload in progress. [4] Download/upload complete. Select/jump complete. Speak output in progress. F600h F6Utt F6 Utt Dst1H Dst1L...DstNH DstNL F6 Utt EminH EminL EmaxH EmaxL F6 Utt Dst1H Dst1L...DstNH DstNL EminH EminL EmaxH EmaxL Aborting SI listen mode. Utt = utterance ID. Utterance ID, high/low byte of distance to utterance 1...utterance N. Utterance ID, high/low byte of min. and max. energy value, Utterance ID, high/low byte of distance to utterance 1...utterance N, high/low byte of minimum energy value, high/low byte of maximum energy value. Trigger detection code (see init command). Rejection: utterance too loud. Rejection: utterance too long. Rejection: utterance begins too soon. Rejection: bad signal/noise ratio. Rejection: reason uncertain. F63Ah F63Bh F63Ch F63Dh F63Eh F63Fh F700h F73Eh F73Fh F740h F341h F7Utt SD Recognition F344h F7Utt DstH DstL F351h F7Utt Dst1H Dst1L... Result DstNH DstNL F361h F7Utt EminH EminL EmaxH EmaxL F371h F7Utt Dst1H Dst1L... DstNH DstNL EminH EminL EmaxH EmaxL Aborting SD Listen mode. After SD utterance search: not found. Rejection. Sort completed. After SD utterance search: empty. Rejection: MSM6679A-110 SD memory full/empty. After SD utterance search: in use. Utt = Utterance ID triggered. Utterance ID, high/low byte of distance. Utterance ID, high/low byte of distance to utterance 1... utterance N. Utterance ID, high/low byte of minimum energy value, maximum energy value. Utterance ID, high and low byte of distance to utterance 1... distance to utterance N, high and low byte of minimum energy value, maximum energy value. Vector Upload F743h 0000h Upload failure. F743h NH NL V1H V1L...VNH VNL High/low bytes of length of vector, V, high/low byte of first V...Nth V. Trap Error Codes F801h F802h F804h F808h F810h F820h F840h F880h Record Response FA00 Reserved. Invalid SP header or table. Select/jump error. Divide-by-zero error. Unclassified download/upload error. Memory full; 32-Kbyte block boundary violation error. Sample data over-run. [1] Invalid message received. Record complete. 29 MSM6679A-110 Voice Recognition Processor 1. 2. 3. 4. 5. 30 ¡ Semiconductor Sample data overrun issued when real-time SP in Listen mode cannot keep up with incoming samples, i.e., if the A/D signal input routine overwrites a sample data buffer before it is fully processed. This acknowledge is sent only if Init command 1111 0010 xxxx x1xx (F2 xxxx x1xx) is set to enable acknowledgments. These messages are sent in response to a request command (F5XYh) from the host. Upload/download in progress, acknowledging load request immediately before data transfer. If in response to an N-byte download request, the MSM6679A-110 then receives N bytes (if N is even, or N+1 if N is odd) of data from the host. If N is odd and N+1 bytes are received, only N bytes are written to MSM6679A-110 memory. If in response to an upload, the MSM6679A-110 then sends N bytes (if N is even, or N+1 if N is odd) of data to the host. If an utterance was recognized, XYh is the utterance identity or class number, and additional parameters may be appended, if requested in the SI Recog (F3XYh with X=0...3) command. Otherwise, XYh indicates various results as detailed. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Command Descriptions Purge Operand F000 Description Return Values Purge MSM6679A-110 Input Stack. This command clears the MSM6679A-110 input stack of commands that are waiting to be executed. Commands already in progress, such as a None pending MSM6654 poll action, are not affected. It does not affect the MSM6679A-110 output stack. Set Parameter Operand F102h XXYYh Description Set SP/SI Recognition Origin. Prior to SD or SI recognition, address pointers must be set to point at the SP or SI recognition parameter tables.This command sets the starting address of SP and SI recognition parameter tables. This address is the location of the first word of a header that contains pointers to one or more individual SP/SI tables. XXYYh = high (XXh) and low (YYh) bytes of requested address. The MSM6679A-110 uses and returns an even address outside the MSM6679A-110 work space that is as near as possible to the requested address. Leave this parameter at its default value unless you are using an Oki custom SI vocabulary and are instructed to alter SP/SI recognition origin. Default SP/SI origin: 8000h Return Values [1] F102h XXYYh = High (XXh) and low (YYh) bytes of resultant address. If a valid header is not found at the resultant address, the MSM6679A-110 immediately sends response code: F802h = Invalid SP/SI header. F103h XXYYh Set SD Recognition Origin [2]. This command sets the SD origin address at the starting address of the current SD recognition parameter table. This command may be used to select among multiple RAM-resident SD vocabulary tables. XXYYh = high (XXh) and low (YYh) bytes of requested address. The MSM6679A-110 uses and returns an even F103h XXYYh = high (XXh) and address outside the MSM6679A-110 work space that is as low (YYh) of resultant address. near as possible to the requested address. Leave this parameter at its default value unless you are using an Oki custom vocabulary and are instructed to alter SD recognition origin. The table length is 0A7Ch bytes. Default SD origin: 4A00h F104h XXYYh Set Triggering Origin. This command sets the starting address of triggering parameter tables. This address is the location of the first word of a section of data memory containing one or more contiguous triggering parameter tables. XXYYh = high (XXh) and low (YYh) bytes of requested F104h XXYYh = high (XXh) and address. The MSM6679A-110 uses and returns an even low (YYh) bytes of resultant address outside the MSM6679A-110 work space that is as address. near as possible to the requested address. Leave this parameter at its default value unless you are using an Oki custom SI vocabulary and are instructed to alter triggering origin. Default triggering origin: F100h. 31 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Set Parameter (Continued) Operand Description Return Values [1] F11Yh Set IRQ Level. This command requests direction of host interrupts to IRQ Y. The MSM6679A-110 then selects IRQ Z, where Z is the nearest legal value to Y. Legal IRQ values are F11Zh = IRQ Z selected. any from the set {5 (default),A,B,C}. Default IRQ level: 5 F12Yh Set SD Recognition SP table. This command sets the SP parameter table number to be used in processing speech input during SD Recognition. The MSM6679A-110 selects SP table number Z, where Z is the nearest valid value to Y. By default, the MSM6679A-110 selects SP table 3 until this command is issued. This command selects SP parameters only, and does not select among multiple RAM-resident SD vocabulary tables, which can be independently selected by the Set SD Origin command (F103h). After setting the table number and returning the resultant value, the MSM6679A-110 checks the validity of the SP header. If the header is invalid, an error message is returned. Set this value to (NSI +1), where NSI is the number of SI subvocabularies. Default SP table: 3. F12Z = SP table Z selected. If the SP header is invalid, a second message follows: F802h = Invalid SP header. F130h VN TN Select Triggering Table. This command selects triggering table TN for use with SP table VN. Valid values for VN and TN are between 01h and 0Fh. Leave this parameter at its default value unless you are using an Oki custom SI vocabulary and are instructed to alter the triggering table. F130h f(VN) f(TN) = Triggering table selected. Default = 0101, 0202, 0303... F440h Set ISA Mode. This command sets the port configuration for the ISA bus. None. Default is off. 1. 2. 32 Return value is actual parameter value which may not equal the set parameter value. See also F6XY ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Initialize F2xx Bit Power-On/ Values Reset Value Action Return Value After power-on, the MSM6679A-110's mode corresponds to that after issuing a F20C command. This mode may NOT be the optimum condition for most situations, so the user is advised to carefully understand the desired condition and develop a suitable command for the application at hand. In addition, ensure that unwanted bits do not get set or reset when attempting to set individual conditions. The conditions selected are based on the XXh values associated with the last F2 command issued. 1xxx xxxx x1xx xxxx Cleared Cleared Background Noise Initialization. When set to 1, the MSM6679A110 starts a 500-ms background noise initialization. When set to 0, the MSM6679A-110 does not perform background noise initialization. The MSM6679A-110 requires this command prior to recognition for noise vector subtraction during the utterance sampling period. Use the background initialization command whenever there is a change in the background noise level. For example, sample the noise signature in a vehicle at rest and moving at 35 MPH with its windows rolled down. The quality of a phone line connection can also vary from call to call. The host MCU must implement a strategy as to when to issue a background initialization command. In a vehicle, the host MCU could monitor the vehicle speed, fan speed, radio volume, etc. Alternatively, the host MCU could issue this command each time a new recognition session starts or a new line connection is established. However, the 0.5-sec sample period could degrade system responsiveness if used too frequently. A zero in this bit location during the F2XXh command will not cause an initialization. The F505h command causes the same initialization sequence. Wait for Recognition Command/Auto Restart SI Recognition. When set to 1, the MSM6679A-110 waits for a recognition command after each response. When set to 0, the MSM6679A110 auto-restarts SI recognition after each response. This bit should be set to 1 when an action is to be taken immediately after an utterance. Auto-restart recognition is the desired mode during digit string recognition, automated tape testing of digits, or in demonstrations where continuous recognition is desired. F501 = Background initialization complete F2XY = Initialization acknowledge. [1] F2XY = Initialization acknowledge. [1] 33 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Initialize (Continued) F2xx Bit Power-On/ Values xx1x xxxx xxxx 1xxx xxxx x1xx xxxx xx1x xxxx xxx1 1. 34 Reset Value Action Return Value Cleared Beep After Each Voice Trigger. When set to 1, the MSM6679A-110 beeps after each voice trigger. When set to 0, the MSM6679A-110 does not beep after each voice trigger. These beeps do not cause a F400h message to be issued to the host MCU. When set to 1, the MSM6679A-110 beep can help a user avoid speaking before the MSM6679A-110 is ready. This mode is normally used with a digits vocabulary to pace the user and confirm each utterance reception. F2XY = Initialization Instead of using beeps, an external MSM665x speech synthesizer acknowledge. [1] can repeat digits as they are recognized. However, some users find the number repetition annoying. Therefore, firmware could repeat digits during initial usage and switch to beep mode later. Typically, performance improves with time as users learns to speak with the correct enunciation and volumes. The MSM6679A-110 in this case trains the user. Note that the host MCU can also make the MSM6679A-110 beep with the F47Eh command. Set Set Output Volume. When set to 1, VOICEOUT1 sound output level is set to half of full volume (80h). When set to 0, voice output level is unaffected. MSM6679A-110 sound output volume can also be set at any level F2XY = Initialization on a continuous scale from 00h to FEh (low to high) with the acknowledge. [1] FEXXh command. The MSM665x speech synthesizer has four discrete sound output volumes, corresponding to 0h - 20h, 21h 40h, 41h - 80h, and 81h - FEh. Set Send Response Code After Sound Output. When set to 1, the MSM6679A-110 issues an acknowledge response (F400h) when sound output is completed. When set to 0, the MSM6679A-110 F2XY = Initialization does not issue an acknowledge response when speech response is acknowledge. [1] completed. Automatic beeps after voice triggers do not cause an F400h command to be issued. Cleared Trigger Detection Only. When set to 1, the MSM6679A-110 does not sort SI vocabularies for the best match, instead returning F63Ah code when an utterance has been detected. When set to 0, normal recognition is performed. When this bit is set to 1, the host MCU can use the F343h command to upload the recognition parameter vector, so that the host can perform independent processing. F2XY = Initialization acknowledge. [1] Cleared Clear SD Recognition and Name Tag RAM. When set to 1, the MSM6679A-110 initializes the SD parameter table. When set to 0, existing SD parameters are preserved. After this bit is set to 1, all SD training and name tag pointers are erased. Use this command to start training for a new user. If the old name tags are to be retained, the F50Ch command can recall old name tags from FLASH. To set up for a blank SD and name tag table at the next power-on, issue the command sequence F201h F507h. F2XY = Initialization acknowledge. [1] See the Response Summary table earlier in this section for a complete description of the XY codes in initialization acknowledgment messages. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Recognize Opcode F300h Action Stop Listening. This command causes the MSM6679A-110 to exit SI or SD Listen mode, whichever was active. Start SI Listen Mode. For all the following opcodes, the MSM6679A-110 performs SI recognition on incoming utterances, using SI vocabulary Y. The vocabulary Y is identified by one of 15 sets, thus Y = 1h ~ Fh. MSM6679A-110 was not in Listen mode. Aborting SI Listen mode. F700h Aborting SD Listen mode. F600h Aborting SI Listen mode. F63Ah Trigger detection code (see Initialization command). F63Bh~F63Fh Rejection. Invalid signal processing table. F840h Sample data overrun. Return recognized phrase using vocabulary number Y. F6h Utt Utterance ID in vocabulary Y. Return recognized phrase and distance table for vocab Y. Utterance ID in vocabulary Y, high and F6h Utt Dst1H Dst1L... low byte of distance to utterance 1... DstNH DstNL distance to utterance N. Return recognized phrase and energy value for vocab Y. Utterance ID in vocabulary Y, high and F6h Utt EminH EminL low byte of minimum and maximum EmaxH EmaxL energy value. Return recognized phrase, distance table, and energy value for vocab Y. F6h Utt Dst1H Dst1L... DstNH DstNL EminH EminL EmaxH EmaxL Utterance ID, high and low byte of distance to utterance 1...distance to utterance N, high and low byte of minimum and maximum energy value. F740 Triggered. F700 Abort SD Listen mode. F73E Rejection. F73F Memory empty. F802 Invalid SP table. F840 Sample data overrun. F73Fh Abnormal response: Memory empty. F341h Return recognized phrase for vocab Y. This command can be issued several times to yield first, second, third best, etc. F7h Utt Utt= Utterance ID. F344h Return recognized phrase and distance for the current vocabulary. F7h Utt DstH DstL Utt = index of recognized phrase, DstH DstL = high/low bytes of distance from nearest phrase. F351 Return recognized phrase and distance table for vocab Y. F7h Utt Utterance ID, high and low byte of Dst1H Dst1L... distance to utt. 1...N. DstNH DstNL F361h Return recognized phrase and energy value for vocab Y. F7h Utt Utterance ID, high and low byte of EminH EminL minimum and maximum energy value. EmaxH EmaxL F301h F33Fh F31Yh F32Yh F33Yh Start SD Listen Mode. When an utterance is captured, it is analyzed and converted to a "recognition parameter vector." The host may then command the MSM6679A-110 to use this vector in various ways (e.g., Sort, Update, or Recognition Vector Upload). SD Recognition Sort. These commands sort the distances between the recognition parameter vector and the reference vectors for the utterances in the current SD vocabulary. F341h, F344h, F351h, F361h, F371h None F600h F802h F30Yh F340h Return Value 35 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Recognize (Continued) Opcode Action F341h, F344h, F351h, F361h, F371h F371h F342h Update SD Recognition Enrollment. This command updates enrollment on utterance Utt, immediately after a "F7h Utt" response to the Sort SD Distances command (F341h). Alternatively, the utterance to be updated can be selected by the SD Search command (F9XYh). This command uses the recognition parameter vector from the most recently captured utterance, and does not start SD Listen mode. Generally, update should be performed only if correct utterance identify is confirmed by the user. F343h Return recognized phrase, distance table, and energy value for vocab Y. Recognition Vector Upload. Request recognition parameter vector upload to host. Return Value F7h Utt Dst1H Dst1L... DstNH DstNL EminH EminL EmaxH EmaxL Utterance ID, high and low byte of distance to utterance 1...distance to utterance N, high and low byte of minimum and maximum energy value. F740h Update complete. F743h NH NL V1H V1L... VNH VNL = Success, where NH/NL = high/low bytes of N, N = Length of recognition parameter vector V, V1H/V1L = high/low bytes of first element of V, VNH/VNL = high/low bytes of Nth element. F743h 00 00 Failure. Speak Opcode Action Speak Phrase from External Memory. This command causes the MSM6679A-110 to play back a name tag from external memory. If no F401h ~ sound is defined for a selected index, the F43Dh MSM6679A-110 plays a beep. See the Record commands for information on creating name tags. Speak Phrase from Low Internal Memory. If no sound is defined for a selected index, the MSM6679A-110 plays a beep. The default phrases supplied with the MSM6679A-110 in the smaller low playback memory area are F441h ~ listed below. F450h F441h Drip. 36 F442h Buzzer. F443h Dial tone. F444h Bonk. Return Value F400h If enabled, this value is returned upon completion of playback. F400h If enabled, this value is returned upon completion of playback. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Speak (Continued) Opcode Action Return Value Speak Phrase from High Internal/External Memory. If no sound is defined for a selected index, the MSM6679A-110 plays a beep. The default phrases supplied with the MSM6679A110 in the larger upper playback memory area are listed below. F451h "0" simulated DTMF tone. F452h "1" simulated DTMF tone. F453h "2" simulated DTMF tone. F451h ~ F454h F47Ch F455h "3" simulated DTMF tone. F456h "5" simulated DTMF tone. F457h "6" simulated DTMF tone. F458h "7" simulated DTMF tone. F459h "8" simulated DTMF tone. F45Ah "9" simulated DTMF tone. F45Bh "*" simulated DTMF tone. F45Ch "#" simulated DTMF tone. F400h If enabled, this value is returned upon completion of playback. "4" simulated DTMF tone. F47D Reserved. This command is reserved for future use. — — F47Eh Beep. This causes the MSM6679A-110 to beep for 50 ms. F400h If enabled, this value is returned upon completion of playback. F47Fh Pause. This command can be issued while the MSM6679A-110 is performing sound output and is then put in the MSM6679A-110 command stack for subsequent processing. F400h When this command is executed, sound output pauses for 0.2 sec. The pause command is useful for word spacing. If enabled, this value is returned upon completion of playback. F480h Set MSM6654 Mode. This command causes the MSM6679A-110 to initialize None. the external MSM665x device, also clearing the device from BUSY mode. Playback Sound from MSM665x Device. This command causes the MSM6679A-110 to issue a speak command to the MSM665x slave device. F481h The value is passed on the MSM665x device as F400h F4FFh 01h - 07Fh. The actual phrase is determined by the vocabulary programmed into the MSM665x device. Up to 127 external phrases are supported. F50Bh Set MSM665x Busy Mode ON. If enabled, this value is returned upon completion of playback. If NAR is set, the F400h command is sent when the MSM665x device is ready for another command. If busy mode is selected, the F400 command is returened when the sound is finished. None. 37 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Speak (Continued) Opcode F51Bh Action Return Value Set 6654 NAR mode. This command, which is the complement of the F50B command, sets up the handshaking to the attached 6654 speech None. synthesizer to use the NAR. This setup uses the 6654's double buffer feature to eliminate any gap between two consecutive phases. Set Output Level. This command sets the speech output level to one of 255 values as follows: FEXYh FE03 Set minimum output level. FE80h Set output level half way (default). FEFEh Set maximum output level. None. Request Opcode F500h Action Status Request. This command causes the MSM6679A-110 to return a 2-byte value indicating its current status. Return Value F500h MSM6679A-110 ready. F520h MSM6679A-110 disabled. F540h MSM6679A-110 waiting for start. F560h MSM6679A-110 waiting for end. F580h MSM6679A-110 processing. F5A0h Download/upload in progress. F5C0h Download/upload complete. F5E0h Select/jump complete. F501h Select last FLASH bank for SI recognition. F510h Select download RAM bank for SI/SP template area. This command enables the download RAM bank in the upper 32 K of data memory for SI recognition. No return value F520h Select buffer RAM bank for SI/SP. This command enables the buffer RAM bank in the upper 32 K of data memory for SI recognition. No return value F522h Copy download RAM bank to buffer RAM bank. This command copies the download RAM bank F501h to the buffer RAM bank. The copied address range is (8000-FFFF). Copy is complete. F513h Save download RAM bank templates in first FLASH. Save the download RAM SI/SP area (8000-F2FF) to the same address range in the first FLASH. Save is complete. 38 F501h ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Request (Continued) Opcode Action F514h Get download RAM bank templates from the first FLASH. Recall the download RAM SI/SP template (8000 - FFFF) from the same address range in the first FLASH. F501h Save is complete. F515h Save download RAM bank templates in last FLASH. SAVE the download RAM bank SI/SP template area (8000 - F2FF) to the same address range in the last FLASH. F501h Save is complete. F516h Get download RAM bank templates from last FLASH. Recall the download RAM bank SI/SP template area (8000 - FFFF) to the same address range in the last FLASH. F501h Save is complete. Download/Upload. Full syntax: F5 02 00 Ctl AdH AdL NH NL [Dt1... DtN [Dt(N+1)]] Full syntax: F5 02 00 Ctl AdH AdL NH NL [Dt1... DtN [Dt(N+1)]] Ctl(7) = 0 for download, Ctl(7) = 1 for upload Ctl(6) = 0 for data RAM, Ctl(6) = 1 for program RAM/ROM If Ctl(6)=0 then Ctl(1-0) = Seg: Data segment selection If Ctl(6)=1 and Ctl(1-0) = x0, then external program segment 0 is used. If Ctl(6)=1 and Ctl(1-0) = x1, then external program segment 1 is used. F502h AdH AdL = high, low bytes of starting address. NH NL = high, low bytes of N N = Number of bytes to be downloaded or uploaded (maximum 07FFCh) Dt1... DtN = Download data. Note (here and in upload response) that data are 8-bit binary values, even if using the serial interface. Dt(N+1). If N is odd, an extra byte is appended to the data so that the total number of bytes in the message remains even. This command requests data transfer to/from data or external program memory.The control parameter (Ctl) controls the direction of the transfer (i.e., download vs. upload) and specifies which of six 64-Kbyte memory segments (i.e., four data segments and two external program segments) is to be accessed. This command does not work with internal program memory. It is not possible to download to external program memory while running in external program memory. The address and length parameters (AdH AdL NH NL) specify the starting address and length of the transfer in bytes. Since the MSM6679A-110 can only perform download /upload transfers within one 32-Kbyte block in one Download /Upload command, the address and length parameters must not specify a transfer that violates a 32-Kbyte address boundary. If this restriction is violated, the download/upload request will be denied. Return Value Immediately after receiving parameter NL, the MSM6679A-110 responds with a message to indicate acceptance or denial of the transfer request. Acceptance is indicated by F5A0h. Denial is indicated by a F8XYh. At the end of an accepted transfer, the MSM6679A-110 responds with a message to confirm or deny valid completion of the transfer. Valid completion is indicated by F5C0h. F880h Invalid message received. F840h Sample data over-run. F820h 32-Kbyte block boundary violation error. F810h Unclassified download/upload error. F808h Divide-by-zero error. F804h Select/jump error. F802h Invalid SP header or table. F801h Reserved. FAXYh Most and least significant byte of address where error occurred. FBXYh 39 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Request (Continued) Opcode Action Return Value Select/Jump. This command selects a new data segment, or Jumps to a new program segment. Ctl(7)=0 is used to first select a new data segment. Ctl(7)=1 then jumps to that program segment. Upper 32-Kbyte of Seg(7)=0 selected segment is accessed normally. Access lower 32-Kbyte Ctl(7)=0 Seg(7)=1 block of selected segment in upper 32 Kbytes of data space. F5E0h Success. F8XYh Failure, with XY(2) = 1. F5E0h Success. F8XYh Failure, with XY(2) = 1. XXXX Four-digit ASCII number. Seg(6~2) Reserved. Seg(1~0) Data segment selection. F503h Ctl Seg Seg(7)=0 Jump to selected external program segment. Seg(7)=1 Jump to internal program segment. Seg(6~1) Reserved. Ctl(7)=1 Seg(0) F504h 40 If Seg(7) =1, not used. If Seg(7) = 0 and Seg(0) = 0: external program segment 0. If Seg(7) = 0 and Seg(0) = 1: external program segment 1. Retrieve MSM6679A-110 Firmware Revision Number. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Request (Continued) Opcode Action Return Value F505h Initialize in Background. Background noise initialization is performed for 500 ms. The MSM6679A-110 requires this command prior to recognition for noise vector subtraction during the utterance sampling period. Use the background initialization command whenever there is a change in the background noise level. For example, sample the noise signature in a vehicle at rest and moving at 35 MPH with its windows rolled down. The quality of a phone line connection can also vary from call to call. The host MCU must implement a strategy as to when to issue a background initialization F501h command. In a vehicle, the host MCU could monitor the vehicle speed, fan speed, radio volume, etc. Alternatively, the host MCU could issue this command each time a new recognition session starts or a new line connection is established. However, the 0.5-sec sample period could degrade system responsiveness if used too frequently. A zero in this bit location during the F2XXh command will not cause an initialization. The F2xxh command can also be used to perform background noise initialization. Initialization is complete. F506h Retrieve Vocabulary and Trigger Table Revision XXXX Number. Four digit ASCII number. F507h Save SDR templates in last FLASH. Save the download RAM bank SD template area. Saves 2684 bytes from the address set by the F103 command to the address range F300FD7F in the last FLASH. The default is 4A00547B→F300-FD7F). F501h Save is complete. F508h Get SDR templates from last FLASH. Get the download RAM bank SD template area. Saves 2684 bytes to the address set by the F103 command from the address range F300FD7B in the last FLASH. The default is (F300FD7B→4A00-547B). No return value F509h Select Default SI Vocabulary. (First FLASH) — — 41 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Record Opcode Action F101h 00XXh Set Name Tag Length, Set MSM665x Busy Mode ON. Name tag record length is set by XXh, with XXh defining record length in 14-ms intervals. The maximum record length of FFh yields a recording interval of 3.57 sec. The default value is 1.2 sec. F101h 00XXh Operation complete. F105 xxxx Set Name Tag Record Origin. This command sets the beginning address for recording name tags. XXXX = 128 byte blocks from 0000 to 02FF. The reset default is 0000. This is only effective before an F50A command since new recordings start after the end of the previous recording. The F50A command uses this number to calculate the first address. F105 BAAA, where B is the bank number (0,1,2), and AAA is the bank address /16 (800 - FF8) F106 xxxx F106 BAAA, where B is the Set Name Tag Record End. This command sets bank number the ending address for recording name tags. (0,1,2), and XXXX = 128 byte blocks from 0000 to 02FF. AAA is the The reset default is 01FF. bank address /16 (800 - FF8) F50Ah Clear Name Tag Table. F50Ch Recall name tag pointers from first FLASH. Save the first FLASH name tag pointers (FD80 F501h FFFF) to the working name tag pointer table. The default is (FD80-FFFF→5480-56FF). Saved name tag table recalled. F51Ch Recall name tag pointers from last FLASH. Save the last FLASH name tag pointesr (FD80 FFFF) to the working name tag pointer table. The default is (FD80-FFFF→5480-56FF). F501h Name tag pointers recalled. F50Dh Save name tag pointers in first FLASH. Save the working name tag pointer table to the first FLASH name tag pointers. The default is (5480 -56FD→FD80-FFFD). F501h Name tag table saved. F51Dh Save name tag pointers in last FLASH. Save the working name tag pointer table to the last F501h FLASH name tag pointers. The default is (5480 -56FD→FD80-FFFD). Name tag pointers saved. F50Eh Set Record Volume HIGH. — — F50Fh Set Record Volume to Normal. This is the default setting. — — FA00h Reserved. This command is reserved for future use. — — FA00h Completed. F280h Memory full. FA01h ~ FA3Dh Record Name Tag. 42 Return Value F501h Name tag table cleared. ¡ Semiconductor MSM6679A-110 Voice Recognition Processor Record (Continued) Opcode Action FA3Dh ~ Reserved. These commands are reserved for FAFFh future use. Return Value — — SD Recognition Control Opcode Action Return Value Recognition performance is largely a function of how well the enrollment data represents subsequent tokens of the enrolled utterances, and performance generally improves steadily with each additional enrollment pass. For most applications, three initial enrollment passes are recommended. Subsequent reference updating can be performed with the SD Recognize Update command (F342). F521h Clear SDR table. This command initializes a blank SD template table. The 2684-byte area from the address set by the F103 command (the working SDR table) is set to zeros. The SDR tables in the FLASH banks are not affected. The default is (4A00 - 547B). F501h F6XYh Set SD Segment Pointer. This command sets the SD segment pointer to XY00h, i.e., set the starting address of the current SD recognition parameter table to XY00h. Issuing this command is equivalent to issuing the Set SD Origin command, F103h XY00h. (For further details of operation, please refer to the description of that command.) No return value. F9XYh Search for SD Utterance XY. This is the first step in adding an utterance to the vocabulary, or in replacing an existing one. The SD vocabulary memory is searched for utt. no. XYh. If it is not found and if sufficient SD memory exists, the MSM6679A prepares to add utterance number XYh to the vocabulary. FB00h FC00h Enroll SD Utterance. This command starts MSM6679A SD Listen mode, then uses the next captured utterance to start or update training of the reference data for SD utterance number XY specified in the most recent Search command (F9XYh). The user must be prompted to say the utterance prior to issuing this command. If the utterance was previously enrolled, a training update is performed; if not, the reference data is initialized. Each utterance in the SD vocabulary must be enrolled at least once before it can be recognized. SDR table is cleared F740h Utterance number found. F700h Utterance number not found. F73Fh Memory full. F740h Operation complete. F700h Aborting SD Listen mode. F73Eh Improper level, must repeat. F802h Invalid signal processing table. F840h Sample data overrun. Erase utterance from SD vocabulary. This command erases the reference parameters for utterance number XYh from the SD vocabulary, F740h where XYh is the utterance number retained from the previous Search command (F9XYh). Operation complete. 43 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Asynchronous Serial Protocol Example All messages to the MSM6679A (except downloads and uploads) are echoed, but replies from the MSM6679A to the host are not echoed by the host. This arrangement facilitates manual communication with the MSM6679A using standard terminals. The following table illustrates the range of MSM6679A functions. Comment Action Voice Input Host MSM6679A Command Response Initialize MSM6679A Host initializes MSM6679A. MSM6679A acknowledges. F258 F258 F200 Install new software ersion. Host requests download to program segment 40, starting at location 0, of 32 Kbytes (7FFCh). MSM6679A accepts request. Host sends 32 Kbytes. (~34 sec at 9600 baud). MSM6679A indicates downloadcomplete. F502 0040 0000 7FFC F5A0 ... 502 0040 0000 7FFC Upload software for verification of transfer. Host requests upload from program segment 0, starting at location 0, of 32 Kbytes (7FFCh). MSM6679A accepts request. MSM6679A sends 32 Kbytes. MSM6679A indicates upload complete. F502 00C0 0000 7FFC F502 00C0 0000 7FFC F5A0 ... F5C0 Run new software. Host commands jump to external program segment 0. MSM6679A begins running new load. F503 8000 F503 8000 F5E0 Load trigger tables at 5000h. Host requests download to data segment 0, starting at location 5000h, of 256 bytes (0100h). MSM6679A accepts request. Host sends 256 bytes (~0.25 sec at 9600 baud). MSM6679A indicates download complete. F502 0000 5000 0100 F502 0000 5000 0100 F5A0 Set new triggering origin. Host requests Set triggering origin to 5000h. MSM6679A sets triggering origin and sends confirming response. F104 5000 F104 5000 F104 5000 Download new SD vocabulary. Host requests download to data segment 0, starting at location 6000h, of 4 Kbytes (1000h). MSM6679A accepts request. Host sends 4 Kbytes (~4.3 sec at 9600 baud) MSM6679A indicates download complete. F502 0000 6000 1000 F502 0000 6000 1000 F5A0 44 F5C0 ... F5C0 ... F5C0 ¡ Semiconductor Comment MSM6679A-110 Voice Recognition Processor Action Voice Input Host MSM6679A Command Response Set new SD tables. Host requests Set SD origin to 6000h. MSM6679A sets SD origin and responds. F103 6000 F103 6000 F103 6000 Download first 4 K of SI vocabulary. Host requests download to data segment 0, starting at location 7000h, of 4k bytes (1000h). MSM6679A accepts request. Host sends 4 Kbytes. MSM6679A indicates download complete. F502 0000 7000 1000 F502 0000 7000 1000 F5A0 Host requests download to data segment 0, starting at location 8000h, of 32k bytes (7FFC). MSM6679A accepts request HOST sends 32 Kbytes. MSM6679A indicates download complete. F502 0000 8000 7FFC Set new SP/SI tables. Host requests Set SP/SI origin = 7000h. MSM6679A sets SP/SI origin and responds. F102 7000 F102 7000 F102 7000 Upload data for diagnostics. Host requests upload from data segment 0, starting at location 300h, of 45 bytes (2Dh). MSM6679A accepts request, signals in progress. MSM6679A sends 46 bytes. MSM6679A indicates upload complete. F502 00A0 0300 002D F502 00A0 0300 002D F5A0 Host requests set SP table 3. MSM6679A selects SP table 3 and confirms. Host initializes MSM6679A. MSM6679A acknowledges. F123 F123 F123 F258 F258 F200 F301 F301 F302 F603 F302 F302 F602 F302 Download last 32 K of SI vocabulary. Set up MSM6679A for SI recognition. SI recognition. ... F5C0 F502 0000 8000 7FFC F5A0 ... F5C0 ... F5C0 Host starts SI recognition, vocabulary 1. "Dial" MSM6679A recognizes utterance 3. Host starts SI recognition, vocabulary 2. "Two" MSM6679A recognizes utterance 2. Host starts SI recognition, vocabulary 2. "Three" MSM6679A recognizes utterance 3. F603 45 MSM6679A-110 Voice Recognition Processor Comment SD enrollment. Action ¡ Semiconductor Voice Input Host starts SI recognition, vocabulary 1. Host MSM6679A Command Response F301 F301 "Store" MSM6679A recognizes utterance 7. Get ready to train SD utterance 1. Memory is empty and ready to train. Pass 1; host sends SD enroll command. F901 FB00 F607 F901 F700 FB00 "John Smith" SD utterance 1 initialized. Pass 2; host sends SD enroll command. FB00 "John Smith" SD utterance 1 updated. Pass 3. Host sends SD enroll command. F740 FB00 F740 FB00 FB00 F740 "John Smith" SD utterance 1 updated. SI recognition of control words. Host starts SI recognition, vocabulary 1. F301 F301 F302 F603 F302 F302 F605 F302 F301 F606 F301 "Dial" MSM6679A recognizes utterance 3. Host starts SI recognition, vocabulary 2. "Five" MSM6679A recognizes utterance 5. Host starts SI recognition, vocabulary 2. "Six" MSM6679A recognizes utterance 6. Host starts SI recognition, vocabulary 1. "Store" MSM6679A recognizes utterance 7. SD enrollment. F607 Host prepares MSM6679A to train SD utterance 2 Memory is empty and ready to train. Pass 1; host sends SD enroll command. F902 F902 FB00 F700 FB00 FB00 F740 FB00 FB00 F740 FB00 "Bill Jones" SD utterance 2 initialized. Pass 2; host sends SD enroll command. "Bill Jones" MSM6679A updates SD utterance 2. Pass 3; host sends SD enroll command. "Bill Jones" MSM6679A signals operation completed. SI recognition of control word. F740 Host starts SI recognition, vocabulary 1. F301 MSM6679A recognizes utterance 11. SD recognition. F301 "Call" F60B Host starts SD recognition. F340 F340 F341 F740 F341 F701 "John Smith" MSM6679A signals trigger OK. Host sends SD sort command. MSM6679A recognizes utterance 1. 46 ¡ Semiconductor Comment Name tag recording. MSM6679A-110 Voice Recognition Processor Action Voice Input Host initiates MSM665x port. Host sets recording length to 1 sec. MSM6679A signals operation complete. Host clears name tag table MSM6679A signals operation complete. Host sets record gain to max. level. Start recording tag one. Host MSM6679A Command Response F480 F101 0047 F50A F50E FA01 F480 F101 0047 F101 0047 F50A F501 F50E FA01 "Jane Doe" MSM6679A signals name tag recording complete. Save name tags to FLASH. Name tags saved. Name tag playback. FA00 F50D F50D F501 FEFF F401 FEFF F401 "Jane Doe" F400 Host sets output volume to mid point. Play MSM6679A internal sound 1. FE80 F442 Play back sound from MSM6654. F49F FE80 F442 "bzzzz" F49F "Completed" Host sets volume to max. level. Host commands play back name tag 1. MSM6679A signals playback OK. Sound playback. The information contained herein can change without notice owing to product and/or technical improvements. Please make sure before using the product that the information you are referring to is up-to-date. The outline of action and examples of application circuits described herein have been chosen as an explanation of the standard action and performance of the product. When you actually plan to use the product, please ensure that the outside conditions are reflected in the actual circuit and assembly designs. OKI assumes no responsibility or liability whatsoever for any failure or unusual or unexpected operation resulting from misuse, neglect, improper installation, repair, alteration or accident, improper handling, or unusual physical or electrical stress including, but not limited to, exposure to parameters outside the specified maximum ratings or operation outside the specified operating range. Neither indemnity against nor license of a third party’s industrial and intellectual property right,etc.is granted by us in connection with the use of product and/or the information and drawings contained herein. No responsibility is assumed by us for any infringement of a third party’s right which may result from the use thereof. When designing your product, please use our product below the specified maximum ratings and within the specified operating ranges, including but not limited to operating voltage, power dissipation, and operating temperature. The products listed in this document are intended for use in general electronics equipment for commercial applications (e.g.,office automation, communication equipment, measurement equipment, consumer electronics, etc.).These products are not authorized for use in any system or application that requires special or enhanced quality and reliability characteristics nor in any system or application where the failure of such system or application may result in the loss or damage of property or death or injury to humans. Such applications include, but are not limited to: traffic control, automotive, safety, aerospace, nuclear power control, and medical, including life support and maintenance. 47 MSM6679A-110 Voice Recognition Processor ¡ Semiconductor Certain parts in this document may need governmental approval before they can be exported to certain countries. The purchaser assumes the responsibility of determining the legality of export of these parts and will take appropriate and necessary steps, at their own expense, for export to another country. Copyright 1997 OKI SEMICONDUCTOR OKI Semiconductor reserves the right to make changes in specifications at anytime and without notice. This information furnished by OKI Semiconductor in this publication is believed to be accurate and reliable. However, no responsibility is assumed by OKI Semiconductor for its use; nor for any infringements of patents or other rights of third parties resulting from its use. No license is granted under any patents or patent rights of OKI. 48