Speech Enhancement

Special Thanks to Nathan Parrish

The terms speech enhancement and speech cleaning properly refer respectively to the improvement in the quality or intelligibility of a speech signal and the reversal of degradations that have corrupted it. In practice however, the two terms are used interchangeably.

The aims of speech cleaning vary according to the application and may include:

  • Improvements in the intelligibility of speech to human listeners.
  • Improvement in the quality of speech that make it more acceptable to human listeners.
  • Modifications to the speech that lead to improved performance of automatic speech or speaker recognition systems.
  • Modifications to the speech so that it may be encoded more effectively for storage or transmission.

Most of the literature on the subject has concentrated on the last three points. The distinction is important because it has been found that some approaches that are effective in improving the apparent quality may actually damage intelligibility. Despite its apparent similarity to that of improving intelligibility, improving the performance of speech recognition systems is significantly different because such systems ignore many of the cues used by humans and, in particular, normally use a coarse spectral representation of the signal from which most voicing information has been removed.