The using of removable module technology when developing VOICECOM makes it supports the needs of different bit streams and applications. Currently 8 to 16bitPCM and GSM6.10 bit stream are supported and also supports modes of keying, monitoring, query and real-time. Query mode in the low-end applications can support up to 1,000 orders in a query task; listening mode can effectively capture the key words and realize non-control input of speech. The key technologies speech recognition involves are as follows:
1. Refusal model of Anti-grammar network
2. Second-time HMM mute compression technology
3. Low bit rate speech coding technology
4. Entropy coding technique
5. Safety Control System
6. Memory-sharing and time-sharing dealing
National patents had been applied for the Low-cost speech recognition engine VoiceCom and the low power consumption, extremely low bit rate speech compression program VioceExpress. The main indicators of speech recognition engine VoiceCom are as follows:
1. 1. Dynamic Random Access Memory (RAM): 2KB (training model) / 6KB (self-adaptive model)
2. Read-only memory (ROM): 2KB/10KB (relate to the hardware platform)
3. Flash memory (FLASH): 200 bytes X Command Number (training model does not require flash)
4. Real-time requirements: 4MIPS/15MIPS (relate to the number of orders)
5. Recognition rate: the fixed model is 96%, self-learning model can be up to 98%
6. Voice Format: 8K PCM sampling, 10-16-bit quantization, 10 ms speech frames.
7. Operating System: LINUX, WIN95/200/NT/XP, Palm OS, Penbox OS, PSOS, VxWorks
8. Support CPU / MCU / DSP: 16-bit MCU, TI DSP C54/C55; AGERE DSP1600/16000; Intel 8086, ARM7, ARM9, Motorola DSP56300.
Indicators of the extremely low bit rate voice compression program VioceExpress are as follows:
1. Dynamic Random Access Memory (RAM): 2KB Variable
2. Read-only memory (ROM): 20KB
3. Decoding speed: 1MIPS
4. Compression rate: 2-6KPBS
Indicators of Voicetune speech controller technical after the adoption of VioceExpress and VoiceCom are:
1. Compression rate: 2.4KPBS
2. Recognition rate: more than 96%
3. Real-time Speed: 4MIPS
4. Support Command: 20
5. Refusal Rate: 5%