Speech encoding system, and speech encoding method
AssigneeMitsubishi Denki Kabushiki Kaisha
InventorsTadashi Yamaura, Hirohisa Tasaki
Life of this patent
Estimating the term…
Abstract
A speech encoding apparatus calculates encoding distortion of a noise-like fixed code vector and multiplies the encoding distortion by a fixed weight corresponding to the noise-like degree of the noise-like fixed code vector, calculates encoding distortion of a non-noise-like fixed code vector and multiplies the encoding distortion by a fixed weight corresponding to the non-noise-like fixed code vector, and selects the fixed excitation code associated with multiplication result with a smaller value.
Claims18
A speech encoding apparatus that includes a sound source information encoder for selecting fixed excitation code for generating synthesized speech using the spectrum envelope information, and that outputs speech code of the input speech, said speech encoding apparatus comprising:
- a first distortion calculator for calculating a first encoding distortion corresponding to a noise-like fixed code vector output by a first fixed excitation codebook for a frame of the input speech;
- a first weight assignor for multiplying the first encoding distortion by a first fixed weight that is preset in accordance with a noise-like degree of noise-like fixed code vectors stored in said first fixed excitation codebook;
- a second distortion calculator for calculating a second encoding distortion corresponding to a non-noise-like fixed code vector output by a second fixed excitation codebook for the frame of the input speech;
- a second weight assignor for multiplying the second encoding distortion by a second fixed weight that is preset in accordance with a noise-like degree of non-noise-like fixed code vectors stored in said second fixed excitation codebook;
- and
- a distortion estimator for determining which of the noise-like and non-noise-like fixed code vector corresponds to the smaller one of the multiplication results obtained by said first weight assignor and by said second weight assignor, and selecting the determined one of the noise-like and non-noise-like fixed code vector to be the fixed excitation code,
- wherein the first fixed weight is different than the second fixed weight, and
- wherein the non-noise-like fixed code vectors stored in the second fixed excitation codebook are collectively more pulse-like, thus having a lesser noise-like degree, than the noise-like fixed code vectors stored in the first excitation codebook.
A speech encoding apparatus that includes a sound source information encoder for selecting fixed excitation code for generating synthesized speech, and that outputs speech code of the input speech, said speech encoding apparatus comprising:
- a first distortion calculator for calculating a first encoding distortion corresponding to a fixed code vector output by a first fixed excitation codebook for a frame of the input speech;
- a first weight assignor for multiplying the first encoding distortion by a first fixed weight that is preset in accordance with a number of the fixed code vectors stored in said first fixed excitation codebook;
- a second distortion calculator for calculating a second encoding distortion corresponding to a fixed code vector output by a second fixed excitation codebook for the frame of the input speech;
- a second weight assignor for multiplying the second encoding distortion by a second fixed weight that is preset in accordance with a number of the fixed code vectors stored in said second fixed excitation codebook;
- and
- a distortion estimator for determining which fixed code vector corresponds to the smaller one of the multiplication results obtained by said first weight assignor and by said second weight assignor, and selecting the determined fixed code vector to be the fixed excitation code,
- wherein the first fixed weight is different than the second fixed weight.
A speech encoding method of selecting fixed excitation code for generating synthesized speech, and outputting speech code of the input speech, said speech encoding method comprising the steps of:
- calculating a first encoding distortion corresponding to a noise-like fixed code vector output by a first fixed excitation codebook for a frame of the input speech, and multiplying the first encoding distortion by a first fixed weight that is preset in accordance with a noise-like degree of noise-like fixed code vectors stored in said first fixed excitation codebook;
- calculating a second encoding distortion corresponding to a non-noise-like fixed code vector output by a second fixed excitation codebook for the frame of the input speech, and multiplying the second encoding distortion by a second fixed weight that is preset in accordance with a noise-like degree of non-noise-like fixed code vectors stored in said second fixed excitation codebook;
- determining which of the noise-like and non-noise-like fixed code vector corresponds to the smaller one of the multiplication results;
- and
- selecting the determined one of the noise-like and non-noise-like fixed code vectors to be the fixed excitation code,
- wherein the first fixed weight is different than the second fixed weight, and
- wherein the non-noise-like fixed code vectors stored in the second fixed excitation codebook are collectively more pulse-like, thus having a lesser noise-like degree, than the noise-like fixed code vectors stored in the first excitation codebook.
A speech encoding method of selecting fixed excitation code for generating synthesized speech, and outputting speech code of the input speech, said speech encoding method comprising, when selecting the fixed excitation code, the steps of:
- calculating a first encoding distortion corresponding to a fixed code vector output by a first fixed excitation codebook for a frame of the input speech, and multiplying the first encoding distortion by a first fixed weight that is preset in accordance with a number of the fixed code vectors stored in said first fixed excitation codebook;
- calculating a second encoding distortion corresponding to a fixed code vector output by a second fixed excitation codebook for the frame of the input speech, and multiplying the second encoding distortion by a second fixed weight that is preset in accordance with a number of the fixed code vectors stored in said second fixed excitation codebook;
- determining which fixed code vector corresponds to a smaller one of the multiplication results;
- and
- selecting the determined fixed code vector to be the fixed excitation code,
- wherein the first fixed weight is different than the second fixed weight.
Dependencies are inferred from the claim text.