ASA 130th Meeting - St. Louis, MO - 1995 Nov 27 .. Dec 01

4pSC17. Speech enhancement based on auditory perception.

Jun Yang

Zlatan Ribic

Viennatone GmbH, Froebelgasse 26-32, A-1164 Wien, Austria

It is well known that formants are more important for speech perception than other details of the spectral envelope. Therefore, for the case of speech (vowels) in noise, improving the salience of the formants by decreasing the energy in the valleys between spectral peaks would logically result in improving intelligibility. Furthermore, perceptually only the first three formants are important in determining the sound that is heard, although the higher formants are necessary to produce sounds of acceptable quality. Based on the above, this paper proposes a scheme to enhance the spectral peaks of the first three formants. In this scheme, the first-order high-pass filter with 1-kHz cutoff frequency is used to pre-emphasize the input speech. Three equalizers in parallel form are designed to enhance the amplitudes of the first three formants that are extracted by the cepstral technique and used as the corresponding resonant frequencies of these equalizers. This proposed scheme is particularly effective in the situation where clean speech can be obtained, such as theaters, concert halls, and classrooms with electrical audio-visual aids, etc. The advantages of this scheme over existing schemes will be discussed, and a demonstration tape will also be played during the presentation.