Speech recognition system and speech recognizing method
AssigneeHonda Motor Co., Ltd.
InventorsKazuhiro Nakadai, Toru Takahashi, Hiroshi Okuno
Life of this patent
Estimating the term…
Abstract
A speech recognition system according to the present invention includes a sound source separating section which separates mixed speeches from multiple sound sources from one another; a mask generating section which generates a soft mask which can take continuous values between 0 and 1 for each frequency spectral component of a separated speech signal using distributions of speech signal and noise against separation reliability of the separated speech signal; and a speech recognizing section which recognizes speeches separated by the sound source separating section using soft masks generated by the mask generating section.
Claims6
A speech recognition system comprising:
- at least one processor which is configured to:
- separate mixed speeches from multiple sound sources from one another;
- generate soft masks for a static feature and a dynamic feature respectively of the speech signals, each soft mask taking continuous values between 0 and 1 for each frequency spectral component of a separated speech signal, wherein the processor generates the soft masks based on distributions of speech against separation reliability and distributions of noise against separation reliability;
- and
- recognize the separated speeches using the soft masks.
A speech recognizing method including the steps of:
- separating mixed speeches from multiple sound sources from one another;
- obtaining separation reliability of a separated speech signal;
- separating speech signal from noise;
- obtaining a distribution of speech signal and a distribution of noise against separation reliability;
- generating soft masks for a static feature and a dynamic feature respectively of the speech signals, each soft mask taking continuous values between 0 and 1 for each frequency spectral component of a separated speech signal, wherein the soft masks are generated based on the distribution of speech signal against separation reliability and the distribution of noise against separation reliability;
- and
- recognizing the separated speeches using the generated soft masks.
Dependencies are inferred from the claim text.