Comparative analysis of the clustering of the LPC coefficients in phoneme recognition
Abstract
Coefficient vectors are taken as points of a n-dimensional space to obtain indications on voice data structure by clustering analysis techniques. The automatic clustering methods include iterative minimax, unsupervised radial clustering and shared near neighbors. The initial experimental data were 25 names spoken by 5 speakers. An autocorrelation algorithm with exponential window was used. The recursive and reflexion coefficient vectors were obtained with the help of the Levinson algorithm. The impossibility of multispeaker vowel recognition without a previous speaker normalization, based on the studied methods, is accepted.
- Publication:
-
NASA STI/Recon Technical Report N
- Pub Date:
- December 1983
- Bibcode:
- 1983STIN...8510264P
- Keywords:
-
- Clumps;
- Phonemes;
- Speech Recognition;
- Vector Spaces;
- Algorithms;
- Autocorrelation;
- Linear Prediction;
- Vowels;
- Communications and Radar