Algebraic Models of Speech Segment Databases
TSD '01 Proceedings of the 4th International Conference on Text, Speech and Dialogue
Advances in Very Low Bit Rate Speech Coding Using Recognition and Synthesis Techniques
TSD '02 Proceedings of the 5th International Conference on Text, Speech and Dialogue
Data driven approaches to speech and language processing
Nonlinear Speech Modeling and Applications
| Hi-index | 0.00 |
Very low bit-rate (VLBR) coding of speech offers the opportunity to test methods of automatic generation of sub-word units. This paper describes two approaches to VLBR coding: the first based on ALISP (Automatic Language Independent Speech Processing) techniques, the second based on syllable segments. Experimental results are reported on a database of one Czech professional speaker. The obtained rates for unit encoding were approximately 135 bps for the former approach and 62 bps for the latter. The quality was evaluated by measuring the logarithmic spectral distortion (computed on LPC-spectra), and in informal listening tests. Possible mutual profits of each technique to the other are discussed.