
ii
Introduction
Voice Solutions in Silicon
ISD-SR3000 PRODUCT FEATURES
Speech Recognition Attributes
Speech recognition processor optimized for command and control applications
All speech recognition processing performed on chip
Supports speaker-independent continuous speech
- Command vocabulary selected from a large (>100k word) dictionary
- Number of commands and voicetags determined by external memory availability
- Multiple topics and finite state grammar supported
- Connected digit recognition with no domain restrictions
Supports speaker defined voicetags
- Phonetic models of user speech input created “on the fly”
- More robust than typical speaker-dependent word models
- Used for phone books and customized commands
Recognition always active
- Allows for voice activation with keyword command
- Push-to-talk option
for
battery applications
Zero power voicetag storage
Hidden Markov Models and triphones used to optimize accuracy while maintaining real-time
recognition
Recognition Processor Attributes
Recognition engine optimized for feature extraction and real-time acoustic model search
Interfaces to μ-Law, A-Law or linear voice CODEC
Serial interface to host microcontroller
Single +5V or +3.3V power supply
Current: 40mA (typical) during active recognition
Package: 100-pin QFP
Temperature range: 0 to +70°C