Contextual spelling correction (CSC) for automatic speech recognition (ASR)
Novel solutions for speech recognition provide contextual spelling correction (CSC) for automatic speech recognition (ASR). Disclosed examples include receiving an audio stream; performing an ASR process on the audio stream to produce an ASR hypothesis; receiving a context list; and, based on at least the ASR hypothesis and the context list, performing spelling correction to produce an output text sequence. A contextual spelling correction (CSC) model is used on top of an ASR model, precluding the need for changing the original ASR model. This permits run-time user customization based on contextual data, even for large-size context lists. Some examples include filtering ASR hypotheses for the audio stream and, based on at least the ASR hypotheses filtering, determining whether to trigger spelling correction for the ASR hypothesis. Some examples include generating text to speech (TTS) audio using preprocessed transcriptions with context phrases to train the CSC model.
1. A method of speech recognition, the method
comprising: receiving an audio stream;
performing an automatic speech recognition (ASR) process on the audio stream to produce an ASR hypothesis as a text sequence;
receiving an initial context list;
performing context filtering, wherein the context filtering comprises narrowing down the
initial context list into a preselected context list based on a user's preference, wherein the context filtering comprises adjusting contribution of a relevance weight and a preference weight, the relevance weight comprising an edit distance between the initial context list and the ASR hypothesis, and the preference weight indicating a frequency of usage of a particular context list item; and
based on at least the ASR hypothesis and the context list, performing spelling correction to produce an output text sequence, wherein performing the spelling correction further comprises:
inputting the text sequence into a text encoder;
inputting the preselected context list into a context encoder; and
passing an output of the text encoder and an output of the context encoder into a decoder.
2. The method of claim 1 , further comprising:
performing an action with the output text sequence, wherein the action is selected from the list consisting of:
generating a transcript of the audio stream, initiating a phone call with a contact identified in the audio stream, generating a message to a contact identified in the audio stream, and responding to a query within the audio stream.
3. The method of claim 1 , wherein the context encoder and the text encoder share a set of parameters.
4. The method of claim 1 , further
comprising: filtering ASR hypotheses for the audio stream;
based on at least the ASR hypotheses filtering, determining whether to trigger spelling correction for the ASR hypothesis;
wherein performing spelling correction to produce the output text sequence comprises: based on at least determining to trigger spelling correction for the ASR
hypothesis, performing spelling correction to produce the output text sequence; and
based on at least determining to not trigger spelling correction for the ASR hypothesis, outputting the ASR hypothesis as the output text sequence.
5. The method of claim 1 , further comprising:
training a contextual spell checker, wherein the training comprises generating text to speech (TTS) audio using preprocessed transcriptions with context phrases.
6. The method of claim 1 , further comprising:
segmenting the audio stream into a plurality of audio segments, wherein the ASR hypothesis comprises a hypothesis of speech in an audio segment of the plurality of audio segments.
7. A system for speech recognition, the system
comprising: a processor; and
a computer-readable medium storing instructions that are operative upon execution by the processor to:
receive an audio stream;
perform an automatic speech recognition (ASR) process on the audio stream to produce an ASR hypothesis as a text sequence;
receive an initial context list;
perform context filtering, wherein the context filtering comprises narrowing down
the initial context list into a preselected context list based on a user's preference, wherein the context filtering comprises adjusting contribution of a relevance weight and a preference weight, the relevance weight comprising an edit distance between the initial context list and the ASR hypothesis, and the preference weight indicating a frequency of usage of a particular context list item; and
based on at least the ASR hypothesis and the context list, perform spelling correction to produce an output text sequence, wherein performing the spelling correction further comprises:
inputting the text sequence into a text encoder;
inputting the preselected context list into a context encoder; and
passing an output of the text encoder and an output of the context encoder into a decoder.
8. The system of claim 7 , wherein the instructions are further operative to: perform an action with the output text sequence, wherein the action is selected from the
list consisting of:
generating a transcript of the audio stream, initiating a phone call with a contact identified in the audio stream, generating a message to a contact identified in the audio stream, and responding to a query within the audio stream.
9. The system of claim 7 , wherein the context encoder and the text
encoder share a set of parameters.
10. The system of claim 7 , wherein the instructions are further operative to: filter ASR hypotheses for the audio stream;
based on at least the ASR hypotheses filtering, determine whether to trigger spelling correction for the ASR hypothesis;
wherein performing spelling correction to produce the output text sequence comprises: based on at least determining to trigger spelling correction for the ASR
hypothesis, performing spelling correction to produce the output text sequence; and
based on at least determining to not trigger spelling correction for the ASR hypothesis, output the ASR hypothesis as the output text sequence.
11. The system of claim 7 , wherein the instructions are further operative to:
train a contextual spell checker, wherein the training comprises generating text to speech (TTS) audio using preprocessed transcriptions with context phrases.
12. The system of claim 7 , wherein the instructions are further operative to: segment the audio stream into a plurality of audio segments,
wherein the ASR hypothesis
comprises a hypothesis of speech in an audio segment of the plurality of audio segments.
13. One or more computer storage devices having computer-executable instructions stored thereon, which, on execution by a computer, cause the computer to perform operations comprising:
receiving an audio stream;
performing an automatic speech recognition (ASR) process on the audio stream to produce an ASR hypothesis as a text sequence;
receiving an initial context list;
performing context filtering, wherein the context filtering comprises narrowing down the
initial context list into a preselected context list based on a user's preference, wherein the context filtering comprises adjusting contribution of a relevance weight and a preference weight, the relevance weight comprising an edit distance between the initial context list and the ASR hypothesis, and the preference weight indicating a frequency of usage of a particular context list item; and
based on at least the ASR hypothesis and the context list, performing spelling correction to produce an output text sequence, wherein performing the spelling correction further comprises:
inputting the text sequence into a text encoder;
inputting the preselected context list into a context encoder; and
passing an output of the text encoder and an output of the context encoder into a decoder.
14. The one or more computer storage devices of claim 13 , wherein the operations further comprise:
performing an operation with the output text sequence, wherein the operation is selected from the list consisting of:
generating a transcript of the audio stream, initiating a phone call with a contact identified in the audio stream, generating a message to a contact identified in the audio stream, and responding to a query within the audio stream.
15. The one or more computer storage devices of claim 13 , wherein
the context encoder and the text encoder share a set of parameters.
16. The one or more computer storage devices of claim 13 , wherein the operations further comprise:
filtering ASR hypotheses for the audio stream;
based on at least the ASR hypotheses filtering, determining whether to trigger spelling correction for the ASR hypothesis;
wherein performing spelling correction to produce the output text sequence comprises: based on at least determining to trigger spelling correction for the ASR
hypothesis, performing spelling correction to produce the output text sequence; and
based on at least determining to not trigger spelling correction for the ASR hypothesis, outputting the ASR hypothesis as the output text sequence.
17. The one or more computer storage devices of claim 13 , wherein the operations further comprise:
training a contextual spell checker, wherein the training comprises generating text to speech (TTS) audio using preprocessed transcriptions with context phrases.
18. The one or more computer storage devices of claim 13 , wherein the operations further comprise:
segmenting the audio stream into a plurality of audio segments, wherein the ASR hypothesis comprises a hypothesis of speech in an audio segment of the plurality of audio segments.