Audio Quality Assessment

The audiopreproc module can provide an assessment of the quality of the audio waveform, to give an indication of its suitability for audio indexing.

Clipping Detection

One of the audio quality assessment modes produces an assessment of the percentage of the audio that is clipped. Even a small amount of clipping is highly damaging to transcription accuracy. The numeric clipping assessment is accompanied by a qualitative description of the audio clipping quality (none, insignificant, minor, moderate, or heavy).

Signal-to-Noise Ratio Calculation

The module can analyze audio levels. The module estimates the Signal-to-Noise Ratio (SNR) for the telephone call as a whole, and estimates a file audio level for the waveform regions classified as speech. The module also indicates when two key speakers in a call have significantly different audio levels, which can reduce transcription accuracy.

When you provide a reasonable silence threshold value, the module can also estimate peak audio levels in the audio. For each gain level recognized, the module produces information on the percentage of audio corresponding to that gain level, and the actual energy level value. The module reports a summary of the maximum difference in decibels between speaker levels. For a good quality waveform where the two speakers speak at a similar gain level, this number can be zero (or at least very low).