Speaker recognition
Abstract
Techniques which have been developed for automatic speaker recognition (ASR) are described. All ASR applications include a database of reference templates for known speakers, with the voice input coming from recitation of a reference phrase. Statistical procedures used to implement pattern recognition for assessing the correctness of the received spectral signature are delineated, including normalization, parameterization, feature extraction, similarity comparison and decision steps. The effectiveness of calculating spectral coefficients to enhance ASR accuracy is noted and several factors which should be considered in the design of an ASR are discussed. Finally, available techniques for discerning between speakers are defined.
- Publication:
-
IEEE ASSP Magazine
- Pub Date:
- October 1986
- Bibcode:
- 1986IASSP...3....4O
- Keywords:
-
- Operators (Personnel);
- Pattern Recognition;
- Speech Recognition;
- Voice Communication;
- Cepstral Analysis;
- Differences;
- Efficiency;
- Human Beings;
- Communications and Radar