Methods, devices and computer-readable storage media for real-time speech recognition
AssigneeBaidu Online Network Technology (Beijing) Co., Ltd.
InventorsXiaoyin Fu, Jinfeng Bai, Zhijie Chen, Mingxin Liang, Xu Chen, Lei Jia
Life of this patent
Estimating the term…
Abstract
Methods, apparatuses, devices and computer-readable storage media for real-time speech recognition are provided. The method includes: based on an input speech signal, obtaining truncating information for truncating a sequence of features of the speech signal; based on the truncating information, truncating the sequence of features into a plurality of subsequences; and for each subsequence in the plurality of subsequences, obtaining a real-time recognition result through attention mechanism.
Claims20
A method for speech recognition, comprising:
- based on an input speech signal, obtaining, by one or more computing devices, truncating information for truncating a sequence of features of the speech signal, comprising:
- obtaining, by the one or more computing devices, spike information related to the speech signal by performing connectionist temporal classification (CTC) on the sequence of features, and determining, by the one or more computing devices, the truncating information based on the spike information;
- based on the truncating information, truncating, by the one or more computing devices, the sequence of features into a plurality of subsequences;
- and
- for each subsequence in the plurality of subsequences, obtaining, by the one or more computing devices, a recognition result through attention mechanism.
An electronic device, comprising:
- one or more processors;
- and
- a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are configured to implement:
- based on an input speech signal, obtain truncating information for truncating a sequence of features of the speech signal;
- based on the truncating information, truncate the sequence of features into a plurality of subsequences;
- and
- for each subsequence in the plurality of subsequences, obtain a recognition result through attention mechanism,
- wherein the one or more processors are configured to obtain the truncating information for truncating the sequence of features of the speech signal by actions of:
- obtaining spike information related to the speech signal by performing connectionist temporal classification (CTC) on the sequence of features;
- and
- determining the truncating information based on the spike information.
A non-transitory computer-readable storage medium having stored therein computer programs that, when executed by a processor, cause a processor to implement:
- based on an input speech signal, obtain truncating information for truncating a sequence of features of the speech signal;
- based on the truncating information, truncate the sequence of features into a plurality of subsequences;
- and
- for each subsequence in the plurality of subsequences, obtain a recognition result through attention mechanism, wherein the storage medium further comprises computer programs that, when executed by a processor, cause a processor to implement obtain the truncating information for truncating the sequence of features of the speech signal by actions of:
- obtaining spike information related to the speech signal by performing connectionist temporal classification (CTC) on the sequence of features;
- and
- determining the truncating information based on the spike information.
Dependencies are inferred from the claim text.