Voicelib: Voice recognition for embedded systems
Voicelib is a modular C-language implementation of various voice processing and voice recognition algorithms. Voicelib is intended for rapid prototyping of various implementations of voice recognition (both discrete-utterance and limited vocabulary continuous speech) for embedded applications.
Background & Objective: Voice recognition in embedded systems (e.g. home entertainment, in-vehicle information systems, personal communication, consumer electronics) is expected to be a major product differentiator in the upcoming decade. Voicelib is designed to allow rapid prototyping of various user interfaces and interaction styles. Voicelib is based on publicly available GPL code to minimize coding effort where possible.
Technical Discussion: Voicelib currently implements the following functions on Linux with both a C-language interface and a shell interface: voice capture via commodity soundcard, voice detection and voice gating; format conversions; FFT and inverse FFT transforms; constant-ratio, mel, and mel-cepstral rebinning; feature-vector extraction; sample reordering and tagging; sample plotting, and orthonormal matching. Modules to perform DTW, DTFW, and Viterbi matching are currently under development.
Contact: William Yerazunis
Technology Area: Spoken Language Interfaces
Modification Date: July 6, 2001
