Voice Activity Detection for Speaker Verification Systems

摘要

This paper describes voice activity detection algorithm for speaker verification systems based on properties of human speech modulation spectrum i.e. rate of power distribution in modulation frequency domain. Based on the fact that power of modulation components of speech is concentrated in a range from 1 to 16 Hz and depends on rate of syllables uttering by a person, a new effective algorithm was proposed and compared to standard energy and cepstral based detectors. Experiments confirmed reliability of proposed detector and its better ability to detect speech in real street noise with high probability for signals with very low SNR in compare to other methods.