Enhanced speech activity detection for mobile telephony

A speech activity detector (SAD) based on decision metrics generated in an adaptive differential pulse code modulation waveform coder is described. The detector functions accurately in low-SNR environments characteristic of cellular and personal communications. After nonspeech intervals are removed, the quality of the reconstructed speech is influenced by SAD decision errors and by the suppression of background noise in nonspeech intervals. The author examines subjectively the effect of the SAD on the perceived quality of the received speech.<
>