Add modified beam search for pruned rnn-t.

[csukuangfj]

Training command

./pruned_transducer_stateless/train.py \
  --world-size 8 \
  --num-epochs 60 \
  --start-epoch 0 \
  --exp-dir pruned_transducer_stateless/exp \
  --full-libri 1 \
  --max-duration 300 \
  --prune-range 5 \
  --lr-factor 5 \
  --lm-scale 0.25

Tensorboard log https://tensorboard.dev/experiment/WKRFY5fYSzaVBHahenpNlA/

Decoding command

for epoch in 42; do
  for avg in 11; do
    ./pruned_transducer_stateless/decode.py \
      --epoch $epoch \
      --avg $avg \
      --exp-dir ./pruned_transducer_stateless/exp \
      --max-duration 100 \
      --decoding-method greedy_search \
      --max-sym-per-frame 1
  done
done

for epoch in 42; do
  for avg in 11; do
    ./pruned_transducer_stateless/decode.py \
      --epoch $epoch \
      --avg $avg \
      --exp-dir ./pruned_transducer_stateless/exp \
      --max-duration 100 \
      --decoding-method modified_beam_search \
      --beam-size 4
  done
done

for epoch in 42; do
  for avg in 11; do
    ./pruned_transducer_stateless/decode.py \
      --epoch $epoch \
      --avg $avg \
      --exp-dir ./pruned_transducer_stateless/exp \
      --max-duration 100 \
      --decoding-method beam_search \
      --beam-size 4
  done
done

Decoding results:

decoding method	test-clean	test-other	comment
greedy search (--max-sym-per-frame 1)	2.62	6.37	--epoch 42 --avg 11 --max-duration 100
greedy search (--max-sym-per-frame 2)	2.62	6.37	--epoch 42 --avg 11 --max-duration 100
greedy search (--max-sym-per-frame 3)	2.62	6.37	--epoch 42 --avg 11 --max-duration 100
modified beam search (--beam-size 4)	2.56	6.27	--epoch 42 --avg 11 --max-duration 100
beam search (--beam-size 4)	2.57	6.27	--epoch 42 --avg 11 --max-duration 100

Note:

1. The model is not trained using modified transducer.
2. By modified beam search, it means it hardcodes --max-sym-per-frame=1 during beam search.
3. The current implementation of beam search is super slow and we recommend using only modified beam search.
4. For the decoding time of test-clean and test-other, see the table listed as follows:

decoding method	test-clean (seconds)	test-other (seconds)
greedy search (--max-sym-per-frame=1)	160	159
greedy search (--max-sym-per-frame=2)	184	177
greedy search (--max-sym-per-frame=3)	210	213
modified beam search (--beam-size 4)	273	269
beam search (--beam-size 4)	2741	2221

Will update RESULTS.md and upload the pre-trained model to hugging face later today.

[pkufool]

Does this model contain extra nn.Linear()?

[csukuangfj]

Does this model contain extra nn.Linear()?

Yes. It uses the code from the master.

[csukuangfj]

A pre-trained model, the decoding logs, and the decoding results are uploaded to
https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless-2022-03-12

[danpovey]

A great feature!