Abstract | ||
---|---|---|
We utilize the inter-frame redundancy with the larger-size super-frame structure to realize ultra low bit rate speech encoding. A new clustering model of speech characteristics is proposed to process effectively the parameters of large super-frames. Based on the model, we present algorithms for ultra low bit rate speech encoding at 600 bps and 300 bps for applications in acoustically harsh environments. At the decoder, a close-loop excitation signal magnitude estimation model is employed to improve the naturalness of synthesized speech. Two prototypes have been realized and evaluated using the DRT tests based on the national standard of China. Both prototypes are able to synthesize high quality of speech with DRT score 88.85 and 81.78 respectively. © 2006 IEEE. |
Year | DOI | Venue |
---|---|---|
2006 | null | ICASSP (1) |
Keywords | Field | DocType |
linear predictive coding,encoding,prototypes,speech processing,clustering algorithms,speech synthesis,decoding,speech coding | Speech processing,Speech synthesis,Speech coding,Computer science,Voice activity detection,Algorithm,PSQM,Speech recognition,Harmonic Vector Excitation Coding,Codec2,Linear predictive coding | Conference |
Volume | Issue | ISSN |
1 | null | null |
ISBN | Citations | PageRank |
1-4244-0469-X | 0 | 0.34 |
References | Authors | |
7 | 2 |