ISD-SR3000

The ISD-SR3000 is a complete embedded speech recognition processor. The ISD-SR3000 consists of a speech recognition engine, a speech compression engine, and a recording function. The ISD-SR3000 hardware includes a parallel RISC/DSP core with an optimized instruction set, a flexible CODEC interface, and a serial host controller interface. The speech recognition engine uses sophisticated Hidden Markov Models (HMMs), which enable recognition of continuous speech and connected digits. An application consists of speaker-independent commands (chosen by the application developer), connected digits and speaker-defined commands. The speaker-defined commands allow users to store and recognize voicetags that can be used for custom commands or name list management. The speaker-defined commands use the HMMs, providing much more robust performance compared to conventional speaker-dependent commands. The speaker-independent commands, the audio prompts, and the speaker-defined voicetags are stored in external memory, allowing for maximum application flexibility. Typical storage requirements are 2kB for each pre-defined command, 2.5kB for each audio prompt, and 3kB for each voicetag, including the model, recording, and data. Application commands are divided into topics (menus), with active vocabulary size governed by the size of the external SRAM. A development system, the ISD-DS3000, is available. The development syste includes tools for compiling and sizing commands and prompts, as well as sample C-code for host control program development.