The 'Neural' Phonetic Typewriter
Computer
Efficient vector quantization of LPC parameters at 24 bits/frame
ICASSP '91 Proceedings of the Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference
Real-time implementation and evaluation of variable rate CELP coders
ICASSP '91 Proceedings of the Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference
An efficient approximation-elimination algorithm for fast-nearest-neighbour search
ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
CELP coding at 4.0 kb/sec and below: improvements to FS-1016
ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Improved phonetically-segmented vector excitation coding at 3.4 kb/s
ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Competitive learning and soft competition for vector quantizerdesign
IEEE Transactions on Signal Processing
| Hi-index | 0.00 |
The aim of the paper is to discuss how to combine efficient representations of both LPC parameters and excitation in CELP coders. For this purpose, we have tried a technique used both for VQ and speech recognition. LPC representations will be based on the quantization properties of Self-organizing Maps (SOM) that show as well good properties to detect spectral transitions, and can be used to select a phonetically driven excitation form. The trajectories in SOM's are exploited looking for the improvement of LPC-based speech coders in three different directions: a) to obtain phonetic classifications that assist speech coders to improve the representation of each specific class; b) to save bits by efficient encoding of the LPC envelope; and c) to provide fast search procedures in VQ. These three topics are used to improve the present quality of CELP coders at 4800 bps. and reduce the bit rate while keeping the quality of the synthetic speech. An improved 2400 bps CELP coder is proposed.