Abstract:

The shortages of shorttime energy based voice activity detection method is analyzed first. Then a shorttime signaltonoise ratio (SNR) estimation method is introduced, and a new adaptive voice activity detection algorithm is proposed based on the design of the adaptive discriminative threshold. Simulations with a lot of Gaussian noisy speech signals ranging from 10dB to 20dB show that the comparatively accurate speech endpoints can be detected by the proposed method in different SNR environments.

Key words: voice activity detection;adaptive;signal to noise ratio;discriminative rules