Speech recognition in noisy environment

The performance of speech recognizers degrades as SNR decreases. The performance "knee" is usually located near an average SNR of about 10dB. However, it is difficult to make quantitative statements about performance since it depends on the type of noise and speech. In most situations experimentation is necessary to quantify performance.

The Speech Recognition library utilizes algorithms that dynamically adapt based on estimates of constant background noises. These algorithms attempt to maintain performance in the noisy environment as much as possible. Noisy backgrounds that are not constant, such as competing speech, are more difficult to handle.

Regards,
Lorin Netsch