automatic-speech-recognition audio

pruned_transducer_stateless5 with Conformer encoder for Ukrainian: https://github.com/proger/icefall/tree/uk

Data Filtering

Tensorboard run

./pruned_transducer_stateless5/train.py \
  --world-size 2 \
  --num-epochs 30 \
  --start-epoch 1 \
  --full-libri 1 \
  --exp-dir pruned_transducer_stateless5/exp-uk-shuf \
  --max-duration 500 \
  --use-fp16 1 \
  --num-encoder-layers 18 \
  --dim-feedforward 1024 \
  --nhead 4 \
  --encoder-dim 256 \
  --decoder-dim 512 \
  --joiner-dim 512 \
  --bpe-model uk/data/lang_bpe_250/bpe.model
./pruned_transducer_stateless5/decode.py \
  --epoch 27 \
  --avg 15 \
  --use-averaged-model True \
  --exp-dir pruned_transducer_stateless5/exp-uk-shuf \
  --decoding-method fast_beam_search \
  --num-encoder-layers 18 \
  --dim-feedforward 1024 \
  --nhead 4 \
  --encoder-dim 256 \
  --decoder-dim 512 \
  --joiner-dim 512 \
  --bpe-model uk/data/lang_bpe_250/bpe.model \
  --lang-dir uk/data/lang_bpe_250