A matlab tool for speech processing, analysis and recognition: Sar-lab

Abstract

Presented work is related to research performed in developing a "smart-room." A smart-room can sense all voice activity within the room and pinpoint the source of the audio signal (speaker). The purpose of this audio sensing is two-fold: to monitor for key words, sentences, or phrases that are flagged as relevant by the monitoring entity as well as separation of all acoustic sources from each other (e.g., background noise from the speakers voice) Crucial requirement in successfully creating such a smart-room is the accurate (in terms of recognition performance), efficient (CPU and memory), and consistent recognition of speech (must work equally well for all speakers; i.e., speaker independent, as well as all acoustic environments). To achieve this goal it becomes necessary to develop tools that enable for advanced research in the area of speech processing, analysis and recognition, specifically in this case wake-up-wordi [WUW] recognition. In developing such a system numerous tests of various system models are necessary. Modules ranging from audio signal processing functions and feature extraction, voice activity detection, pattern classification, scoring algorithms, etc., must be combined in order to perform speech recognition. Thus, a major hurdle in this area of research is the analysis, testing, verification, and integration of the individual functions required for speech recognition. To address the analysis and testing issue an appropriate software tool is developed using MATLAB environment that enabled unified framework for tracking the performance of all necessary functions of WUW recognition system. This framework can also be used for testing algorithms and other software components performing speech analysis and recognition tasks. In addition to integrating all of the various components, testing environment can produce additional analysis data all appropriately presented as graphs, charts or images (e.g., spectrogram) that are useful when analyzing and/or troubleshooting such components that are under research. This testing environment has proven to be very useful in aiding research in development of "wake-up word" recognition technology. This tool thus has made research process much more efficient, accurate, and productive. © American Society for Engineering Education, 2006.