Achieving good speech recognition performance: speech audio quality

When using a speech recognition system, one of the critical factors in getting good recognition accuracy is the clarity and fidelity of speech received by the computer. The speech clarity is reduced when there is background noise present near the speaker. Many noise cancelling solutions have been tried to counter this problem, such as the use of DSP technology . The main flaw with many such systems is that they reduce the noise at the expense of the fidelity of the speech, since speech and noise inherently overlap at many frequencies, especially when the noise is competing speech from other speakers in the vicinity.

This is because such systems attempt to remove the noise from the audio signal AFTER the noise and speech are already mixed together. If we could somehow remove the noise before it even gets into the signal, that would enable us to preserve all the critical features of speech, while eliminating the noise, thus sending an optimal audio signal to the speech recognizer. Such a technology has been developed and patented by UmeVoice, and is found in its line of noise cancelling headsets and microphones.