Real-time implementation and evaluation of an adaptive silence deletion algorithm for speech compression

The authors describe a real-time implementation of a modified version of an earlier silence deletion algorithm. The modifications enable real-time implementation using a TMS320C25 digital signal processor and associated hardware, and up to 50 percent compression without significant loss of playback signal quality for single-speaker continuous speech. The algorithm relies on speech input amplitude level, with two adaptive amplitude thresholds, and on the zero-crossing rate. The algorithm operated on 8-bit μ-law encoded samples at 8000 samples/s. Following silence deletion, the remaining samples were left either in μ-law coded form, or were further coded using 4-bit adaptive differential pulse code modulation (ADPCM) for increased compression. This encoding and subsequent decoding prior to playback did not noticeably further degrade the speech quality. Use of silence deletion together with 4-bit ADPCM enabled real-time speech transmission at rates below 19.2 kbit/s