| Summary: | 碩士 === 元智大學 === 通訊工程學系 === 100 === Inside the booming cloud services, speech recognition is one of the most popular features that we know. Microphone array, one powerful enhancement to the benefit of speech recognition, is however not widely applied. Which means there is must more efforts to do to make it acceptable on the market. On this point, I analyzed the computing efficiency of a variety of noise-suppressing combination as well as word-error-rate. In this thesis, a start with using PortAudio API to capture array signal from an ASIO audio device describes the audio interface, and follows with a chapter of post processes including Delay-and-Sum beamforming and two Spectral Subtraction filtering, approached with average-noise estimation and Wiener Filter noise estimation respectively. To build up the analysis basis, HMM-based recognition rate and SNR are included. As the final result, an index, improving rate per unit time, is involved to conduct the results of speech recognition and of filtering-time measurement.
|
|---|
