Rare word recognition with LM-aware MWER training
A method includes generating, using an audio encoder, a higher-order feature representation for each acoustic frame in a sequence of acoustic frames; generating, using a decoder, based on the higher-order feature representation, a plurality of speech recognition hypotheses, each hypotheses corresponding to a candidate transcription of an utterance and having an associated first likelihood score; generating, using an external language model, for each speech recognition hypothesis, a second likelihood score; determining, using a learnable fusion module, for each speech recognition hypothesis, a set of fusion weights based on the higher-order feature representation and the speech recognition hypothesis; and generating, using the learnable fusion module, for each speech recognition hypothesis, a third likelihood score based on the first likelihood score, the second likelihood score, and the set of fusion weights, the audio encoder and decoder trained using minimum additive error rate training in the presence of the external language model.
1. A computer-implemented method executed by data processing hardware that causes the data processing hardware to perform operations comprising:
receiving, as input, a sequence of acoustic frames corresponding to an utterance;
generating, using an audio encoder, at each of a plurality of output steps, a higher-order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
generating, using a decoder, based on the higher-order feature representation at each of the plurality of output steps, a plurality of speech recognition hypotheses, each speech recognition hypotheses corresponding to a candidate transcription of the utterance and having an associated first likelihood score;
generating, using an external language model, for each speech recognition hypothesis of the plurality of speech recognition hypotheses, a second likelihood score associated with the speech recognition hypothesis;
determining, using a learnable fusion module, for each speech recognition hypothesis of the plurality of speech recognition hypotheses, a set of fusion weights based on the higher-order feature representation and the speech recognition hypothesis; and
generating, using the learnable fusion module, for each speech recognition hypothesis of the plurality of speech recognition hypotheses, a third likelihood score based on the first likelihood score, the second likelihood score, and the set of fusion weights,
wherein the audio encoder is trained using minimum additive error rate training in the presence of the external language model by adjusting likelihood scores according to minimum additive error rates among a top-k speech recognition hypotheses output from the decoder.
2. The computer-implemented method of claim 1 , wherein the operations further comprise selecting a speech recognition hypothesis of the plurality of speech recognition hypotheses as a speech recognition result for the sequence of acoustic frames based on the third likelihood scores for the plurality of speech recognition hypotheses.
3. The computer-implemented method of claim 1 , wherein the first likelihood score comprises a conditional likelihood score and an internal language model likelihood score.
4. The computer-implemented method of claim 1 , wherein:
the first likelihood score for a particular speech recognition hypothesis comprises a first plurality of language model scores, each language model score of the first plurality of language model scores corresponding to a particular token of the particular speech recognition hypothesis;
the second likelihood score for the particular speech recognition hypothesis comprises a second plurality of language model scores, each language model score of the second plurality of language model scores corresponding to a particular token of the particular speech recognition hypothesis;
the set of fusion weights for the particular speech recognition hypothesis comprises a first plurality of fusion weights and a second plurality of fusion weights, each fusion weight of the first plurality of fusion weights corresponding to a particular token of the particular speech recognition hypothesis, and each fusion weight of the second plurality of fusion weights corresponding to a particular token of the particular speech recognition hypothesis; and
generating, using the learnable fusion module, the third likelihood score for the particular speech recognition hypothesis comprises:
determining an internal language model score based on the first plurality of language model scores and the first plurality of fusion weights;
determining an external language model score based on the second plurality of language model scores and the second plurality of fusion weights; and
combining the internal language model score and the external language model score.
5. The computer-implemented method of claim 1 , wherein the audio encoder, the decoder, and the learnable fusion module are jointed trained.
6. The computer-implemented method of claim 1 , wherein the external language model is trained on text-only data.
7. The computer-implemented method of claim 1 , wherein the external language model comprises a neural network language model comprising a stack of conformer layers or transformer layers.
8. The computer-implemented method of claim 1 , wherein the external language model comprises a neural network language model comprising a stack of long short-term memory (LSTM) layers.
9. The computer-implemented method of claim 1 , wherein the audio encoder comprises:
a causal encoder comprising an initial stack of multi-head attention layers; and
a non-causal encoder comprising a final stack of multi-head attention layers overlain on the initial stack of multi-head attention layers.
10. The computer-implemented method of claim 1 , wherein the learnable fusion module is configured to compute per-token fusion weights for internal language model scores of the decoder and external language model scores of the external language model.
11. The computer-implemented method of claim 1 , wherein a minimum additive error rate comprises a minimum word error rate.
12. The computer-implemented method of claim 1 , wherein:
the external language model is trained on text-only data; and
the audio encoder and the decoder are trained using hybrid autoregressive transducer (HAT) factorization to facilitate integration of the external language model trained on the text-only data.
13. An automated speech recognition (ASR) system comprising:
an audio encoder configured to:
receive, as input, a sequence of acoustic frames corresponding to an utterance; and
generate, at each of a plurality of output steps, a higher-order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
a decoder configured to:
receive, as input, the higher-order feature representation generated by the audio encoder at each of the plurality of output steps; and
generate, based on the higher-order feature representation received at each of the plurality of output steps, a plurality of speech recognition hypotheses, each speech recognition hypotheses corresponding to a candidate transcription of the utterance and having an associated first likelihood score;
an external language model configured to:
receive, as input, the plurality of speech recognition hypotheses generated by the decoder; and
generate, for each speech recognition hypothesis of the plurality of speech recognition hypotheses, a second likelihood score associated with the speech recognition hypothesis; and
a learnable fusion module configured to, for each speech recognition hypothesis of the plurality of speech recognition hypotheses:
receive, as input, the higher-order feature representation generated by the audio encoder at each of the plurality of output steps;
determine a set of fusion weights based on the higher-order feature representation and the speech recognition hypothesis; and
generate a third likelihood score based on the first likelihood score, the second likelihood score, and the set of fusion weights,
wherein the audio encoder is trained using minimum additive error rate training in the presence of the external language model by adjusting likelihood scores according to minimum additive error rates among a top-k speech recognition hypotheses output from the decoder.
14. The ASR system of claim 13 , wherein the operations further comprise selecting a speech recognition hypothesis of the plurality of speech recognition hypotheses as a speech recognition result for the sequence of acoustic frames based on the third likelihood scores for the plurality of speech recognition hypotheses.
15. The ASR system of claim 13 , wherein the first likelihood score comprises a conditional likelihood score and an internal language model likelihood score.
16. The ASR system of claim 13 , wherein:
the first likelihood score for a particular speech recognition hypothesis comprises a first plurality of language model scores, each language model score of the first plurality of language model scores corresponding to a particular token of the particular speech recognition hypothesis;
the second likelihood score for the particular speech recognition hypothesis comprises a second plurality of language model scores, each language model score of the second plurality of language model scores corresponding to a particular token of the particular speech recognition hypothesis;
the set of fusion weights for the particular speech recognition hypothesis comprises a first plurality of fusion weights and a second plurality of fusion weights, each fusion weight of the first plurality of fusion weights corresponding to a particular token of the particular speech recognition hypothesis, and each fusion weight of the second plurality of fusion weights corresponding to a particular token of the particular speech recognition hypothesis; and
the learnable fusion module is configured to generate the third likelihood score for the particular speech recognition hypothesis by:
determining an internal language model score based on the first plurality of language model scores and the first plurality of fusion weights;
determining an external language model score based on the second plurality of language model scores and the second plurality of fusion weights; and
combining the internal language model score and the external language model score.
17. The ASR system of claim 13 , wherein the audio encoder, the decoder, and the learnable fusion module are jointed trained.
18. The ASR system of claim 13 , wherein the external language model is trained on text-only data.
19. The ASR system of claim 13 , wherein the external language model comprises a neural network language model comprising a stack of conformer layers or transformer layers.
20. The ASR system of claim 13 , wherein the external language model comprises a neural network language model comprising a stack of long short-term memory (LSTM) layers.
21. The ASR system of claim 13 , wherein the audio encoder comprises:
a causal encoder comprising an initial stack of multi-head attention layers; and
a non-causal encoder comprising a final stack of multi-head attention layers overlain on the initial stack of multi-head attention layers.
22. The ASR system of claim 13 , wherein the learnable fusion module is configured to compute per-token fusion weights for internal language model scores of the decoder and external language model scores of the external language model.
23. The ASR system of claim 13 , wherein a minimum additive error rate comprises a minimum word error rate.
24. The ASR system of claim 13 , wherein:
the external language model is trained on text-only data; and
the audio encoder and the decoder are trained using hybrid autoregressive transducer (HAT) factorization to facilitate integration of the external language model trained on the text-only data.