File size: 5,018 Bytes
c4c0def |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 |
---
language: "en"
tags:
- icefall
- k2
- transducer
- aishell
- ASR
- stateless transducer
- PyTorch
license: "apache-2.0"
datasets:
- aishell
metrics:
- WER
---
# Introduction
This repo contains pre-trained model using
<https://github.com/k2-fsa/icefall/pull/219>.
It is trained on [AIShell](https://www.openslr.org/33/) dataset
using modified transducer from [optimized_transducer](https://github.com/csukuangfj/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](https://ieeexplore.ieee.org/document/9054419).
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][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:
```bash
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
```bash
# 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][log]).
The WER for the test dataset is
| | test |comment |
|------------------------|------|----------------------------------------------------------------|
| greedy search | 5.22 |--epoch 64, --avg 33, --max-duration 100, --max-sym-per-frame 1 |
| modified beam search | 5.02 |--epoch 64, --avg 33, --max-duration 100 --beam-size 4 |
# File description
- [log][log], this directory contains the decoding log and decoding results
- [test_wavs][test_wavs], this directory contains wave files for testing the pre-trained model
- [data][data], this directory contains files generated by [prepare.sh][prepare]
- [exp][exp], this directory contains only one file: `preprained.pt`
`exp/pretrained.pt` is generated by the following command:
```bash
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:
```bash
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`.
[icefall]: https://github.com/k2-fsa/icefall
[prepare]: https://github.com/k2-fsa/icefall/blob/master/egs/aishell/ASR/prepare.sh
[exp]: https://huggingface.co/csukuangfj/icefall-aishell-transducer-stateless-modified-2022-03-01/tree/main/exp
[data]: https://huggingface.co/csukuangfj/icefall-aishell-transducer-stateless-modified-2022-03-01/tree/main/data
[test_wavs]: https://huggingface.co/csukuangfj/icefall-aishell-transducer-stateless-modified-2022-03-01/tree/main/test_wavs
[log]: https://huggingface.co/csukuangfj/icefall-aishell-transducer-stateless-modified-2022-03-01/tree/main/log
[icefall]: https://github.com/k2-fsa/icefall
|