icefall-aishell-transducer-stateless-modified-2022-03-01

Introduction

This repo contains pre-trained model using https://github.com/k2-fsa/icefall/pull/219.

It is trained on AIShell dataset using modified transducer from optimized_transducer.

How to clone this repo

sudo apt-get install git-lfs
git clone https://huggingface.co/csukuangfj/icefall-aishell-transducer-stateless-modified-2022-03-01

cd icefall-aishell-transducer-stateless-modified-2022-03-01
git lfs pull

Catuion: You have to run git lfs pull. Otherwise, you will be SAD later.

The model in this repo is trained using the commit TODO.

You can use

git clone https://github.com/k2-fsa/icefall
cd icefall
git checkout TODO

to download icefall.

You can find the model information by visiting https://github.com/k2-fsa/icefall/blob/TODO/egs/aishell/ASR/transducer_stateless_modified/train.py#L232.

In short, the encoder is a Conformer model with 8 heads, 12 encoder layers, 512-dim attention, 2048-dim feedforward; the decoder contains a 512-dim embedding layer and a Conv1d with kernel size 2.

The decoder architecture is modified from Rnn-Transducer with Stateless Prediction Network. A Conv1d layer is placed right after the input embedding layer.

---

Description

This repo provides pre-trained transducer Conformer model for the AIShell dataset using icefall. There are no RNNs in the decoder. The decoder is stateless and contains only an embedding layer and a Conv1d.

The commands for training are:

cd egs/aishell/ASR
./prepare.sh --stop-stage 6

export CUDA_VISIBLE_DEVICES="0,1,2"

./transducer_stateless_modified/train.py \
  --world-size 3 \
  --num-epochs 90 \
  --start-epoch 0 \
  --exp-dir transducer_stateless_modified/exp-4 \
  --max-duration 250 \
  --lr-factor 2.0 \
  --context-size 2 \
  --modified-transducer-prob 0.25

The tensorboard training log can be found at https://tensorboard.dev/experiment/C27M8YxRQCa1t2XglTqlWg

The commands for decoding are

# greedy search
for epoch in 64; do
  for avg in 33; do
  ./transducer_stateless_modified-2/decode.py \
    --epoch $epoch \
    --avg $avg \
    --exp-dir transducer_stateless_modified/exp-4 \
    --max-duration 100 \
    --context-size 2 \
    --decoding-method greedy_search \
    --max-sym-per-frame 1
  done
done

# modified beam search
for epoch in 64; do
  for avg in 33; do
    ./transducer_stateless_modified/decode.py \
    --epoch $epoch \
    --avg $avg \
    --exp-dir transducer_stateless_modified/exp-4 \
    --max-duration 100 \
    --context-size 2 \
    --decoding-method modified_beam_search \
    --beam-size 4
  done
done

You can find the decoding log for the above command in this repo (in the folder log).

The WER for the test dataset is

File description

• log, this directory contains the decoding log and decoding results
• test_wavs, this directory contains wave files for testing the pre-trained model
• data, this directory contains files generated by prepare.sh
• exp, this directory contains only one file: preprained.pt

exp/pretrained.pt is generated by the following command:

epoch=64
avg=33

./transducer_stateless_modified/export.py \
  --exp-dir ./transducer_stateless_modified/exp-4 \
  --lang-dir ./data/lang_char \
  --epoch $epoch \
  --avg $avg

HINT: To use pretrained.pt to compute the WER for the test dataset, just do the following:

cp icefall-aishell-transducer-stateless-modified-2022-03-01/exp/pretrained.pt \
  /path/to/icefall/egs/aishell/ASR/transducer_stateless_modified/exp/epoch-999.pt

and pass --epoch 999 --avg 1 to transducer_stateless_modified/decode.py.