Unified cascaded encoder ASR model for dynamic model sizes
An automated speech recognition (ASR) model includes a first encoder, a first encoder, a second encoder, and a second decoder. The first encoder receives, as input, a sequence of acoustic frames, and generates, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The first decoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a first probability distribution over possible speech recognition hypotheses. The second encoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a second higher order feature representation for a corresponding first higher order feature frame. The second decoder receives, as input, the second higher order feature representation generated by the second encoder, and generates a second probability distribution over possible speech recognition hypotheses.
1. An automated speech recognition (ASR) model comprising:
a first encoder configured to:
receive, as input, a sequence of acoustic frames; and
generate, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
a first decoder configured to:
receive, as input, the first higher order feature representation generated by the first encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a first probability distribution over possible speech recognition hypotheses;
a second encoder configured to:
receive, as input, the first higher order feature representation generated by the first encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a second higher order feature representation for a corresponding first higher order feature frame; and
a second decoder configured to:
receive, as input, the second higher order feature representation generated by the second encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a second probability distribution over possible speech recognition hypotheses,
wherein the first decoder and the second decoder each comprise a respective recurrent neural network-transducer (RNN-T) architecture having a same number of parameters.
2. The ASR model of claim 1 , wherein the first decoder is further configured to generate partial speech recognition results based on the first probability distribution over possible speech recognition hypotheses.
3. The ASR model of claim 1 , wherein the first encoder comprises a causal encoder comprising one of:
a plurality of unidirectional long short-term memory (LSTM) layers;
a plurality of conformer layers; or
a plurality of transformer layers.
4. The ASR model of claim 1 , wherein the second encoder comprises a non-causal encoder comprising one of:
a plurality of unidirectional long short-term memory (LSTM) layers;
a plurality of conformer layers; or
a plurality of transformer layers.
5. The ASR model of claim 1 , wherein the first decoder comprises:
a prediction network configured to:
receive, as input, a sequence of non-blank symbols output by a final softmax layer; and
generate, at each of the plurality of output steps, a dense representation; and
a joint network configured to:
receive, as input, the dense representation generated by the prediction network at each of the plurality of output steps and the first higher order feature representation generated by the first encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, the first probability distribution over possible speech recognition hypotheses.
6. The ASR model of claim 5 , wherein the prediction network comprises:
a long short-term memory (LSTM)-based prediction network; or
a V2 embedding look-up table.
7. The ASR model of claim 1 , wherein the second decoder comprises:
a prediction network configured to:
receive, as input, a sequence of non-blank symbols output by a final softmax layer; and
generate, at each of the plurality of output steps, a dense representation; and
a joint network configured to:
receive, as input, the dense representation generated by the prediction network at each of the plurality of output steps and the second higher order feature representation generated by the second encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, the second probability distribution over possible speech recognition hypotheses.
8. The ASR model of claim 7 , wherein the prediction network comprises:
a long short-term memory (LSTM)-based prediction network; or
a V2 embedding look-up table.
9. The ASR model of claim 1 , wherein the first encoder comprises a greater number of parameters than the second encoder.
10. An automated speech recognition (ASR) model comprising:
a first encoder configured to:
receive, as input, a sequence of acoustic frames; and
generate, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
a first decoder configured to:
receive, as input, the first higher order feature representation generated by the first encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a first probability distribution over possible speech recognition hypotheses;
a second encoder configured to:
receive, as input, the first higher order feature representation generated by the first encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a second higher order feature representation for a corresponding first higher order feature frame;
a second decoder configured to:
receive, as input, the second higher order feature representation generated by the second encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a second probability distribution over possible speech recognition hypotheses; and
a third encoder configured to:
receive, as input, the second higher order feature representation generated by the second encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a third higher order feature representation for a corresponding second higher order feature representation; and
a third decoder configured to:
receive, as input, the third higher order feature representation generated by the third encoder at each of the plurality of output steps; and
generate, at each of the plurality of output steps, a third probability distribution over possible speech recognition hypotheses.
11. A computer-implemented method that when executed on data processing hardware causes the data processing hardware to perform operations comprising:
receiving a sequence of acoustic frames; and
generating, by a first encoder, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
generating, by a second encoder, at each of the plurality of output steps, a second higher order feature representation for a corresponding first higher order feature representation;
at each of the plurality of output steps, generating, by a first prediction network of a first decoder based on a sequence of non-blank symbols output by a final softmax layer, a dense representation;
generating, by a first joint network of the first decoder, at each of the plurality of output steps, a first probability distribution over possible speech recognition hypotheses based on the dense representation generated by the first prediction network of the first decoder; and
generating, by a second decoder, at each of the plurality of output steps, a second probability distribution over possible speech recognition hypotheses,
wherein the first decoder and the second decoder each comprise a respective recurrent neural network-transducer (RNN-T) architecture having a same number of parameters.
12. The computer-implemented method of claim 11 , wherein the operations further comprise generating partial speech recognition results based on the first probability distribution over possible speech recognition hypotheses.
13. The computer-implemented method of claim 11 , wherein the first encoder comprises a causal encoder comprising one of:
a plurality of unidirectional long short-term memory (LSTM) layers;
a plurality of conformer layers; or
a plurality of transformer layers.
14. The computer-implemented method of claim 11 , wherein the second encoder comprises a non-causal encoder comprising one of:
a plurality of unidirectional long short-term memory (LSTM) layers;
a plurality of conformer layers; or
a plurality of transformer layers.
15. The computer-implemented method of claim 11 , wherein the first prediction network of the first decoder comprises:
a long short-term memory (LSTM)-based prediction network; or
a V2 embedding look-up table.
16. The computer-implemented method of claim 11 , wherein the operations further comprise, at each of the plurality of output steps:
generating, by a second prediction network of the second decoder, at each of the plurality of output steps, a dense representation;
receiving, as input to a second joint network of the second decoder, the dense representation generated by the second prediction network at each of the plurality of output steps and the second higher order feature representation generated by the second encoder at each of the plurality of output steps; and
generating, by the second joint network of the second decoder, at each of the plurality of output steps, the second probability distribution over possible speech recognition hypotheses.
17. The computer-implemented method of claim 16 , wherein the second prediction network of the second decoder comprises:
a long short-term memory (LSTM)-based prediction network; or
a V2 embedding look-up table.
18. The computer-implemented method of claim 11 , wherein the first encoder comprises a greater number of parameters than the second encoder.
19. A computer-implemented method that when executed on data processing hardware causes the data processing hardware to perform operations comprising:
receiving a sequence of acoustic frames;
generating, by a first encoder, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames;
generating, by a second encoder, at each of the plurality of output steps, a second higher order feature representation for a corresponding first higher order feature representation;
generating, by a first decoder, at each of the plurality of output steps, a first probability distribution over possible speech recognition hypotheses;
generating, by a second decoder, at each of the plurality of output steps, a second probability distribution over possible speech recognition hypotheses;
receiving, as input to a third encoder, the second higher order feature representation generated by the second encoder at each of the plurality of output steps;
generating, by the third encoder, at each of the plurality of output steps, a third higher order feature representation for a corresponding second higher order feature representation;
receiving, as input to a third decoder, the third higher order feature representation generated by the third encoder at each of the plurality of output steps; and
generating, by the third decoder, at each of the plurality of output steps, a third probability distribution over possible speech recognition hypotheses.