luomingshuang
commited on
Commit
•
073f061
1
Parent(s):
695b5a0
add README.md
Browse files
README.md
ADDED
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Note: This recipe is trained with the codes from this PR https://github.com/k2-fsa/icefall/pull/399
|
2 |
+
# Pre-trained Transducer-Stateless5 models for the AISHELL4 dataset with icefall.
|
3 |
+
The model was trained on the far data of [AISHELL4](https://www.openslr.org/111) with the scripts in [icefall](https://github.com/k2-fsa/icefall) based on the latest version k2.
|
4 |
+
## Training procedure
|
5 |
+
The main repositories are list below, we will update the training and decoding scripts with the update of version.
|
6 |
+
k2: https://github.com/k2-fsa/k2
|
7 |
+
icefall: https://github.com/k2-fsa/icefall
|
8 |
+
lhotse: https://github.com/lhotse-speech/lhotse
|
9 |
+
* Install k2 and lhotse, k2 installation guide refers to https://k2.readthedocs.io/en/latest/installation/index.html, lhotse refers to https://lhotse.readthedocs.io/en/latest/getting-started.html#installation. I think the latest version would be ok. And please also install the requirements listed in icefall.
|
10 |
+
* Clone icefall(https://github.com/k2-fsa/icefall) and check to the commit showed above.
|
11 |
+
```
|
12 |
+
git clone https://github.com/k2-fsa/icefall
|
13 |
+
cd icefall
|
14 |
+
```
|
15 |
+
* Preparing data.
|
16 |
+
```
|
17 |
+
cd egs/aishell4/ASR
|
18 |
+
bash ./prepare.sh
|
19 |
+
```
|
20 |
+
* Training
|
21 |
+
```
|
22 |
+
export CUDA_VISIBLE_DEVICES="0,1,2,3"
|
23 |
+
./pruned_transducer_stateless5/train.py \
|
24 |
+
--world-size 4 \
|
25 |
+
--num-epochs 30 \
|
26 |
+
--start-epoch 1 \
|
27 |
+
--exp-dir pruned_transducer_stateless5/exp \
|
28 |
+
--lang-dir data/lang_char \
|
29 |
+
--max-duration 220
|
30 |
+
```
|
31 |
+
## Evaluation results
|
32 |
+
The decoding results (CER%) on AISHELL4(test) are listed below:
|
33 |
+
When use-averaged-model=False, the CERs are
|
34 |
+
| | test | comment |
|
35 |
+
|------------------------------------|------------|------------------------------------------|
|
36 |
+
| greedy search | 30.05 | --epoch 30, --avg 25, --max-duration 800 |
|
37 |
+
| modified beam search (beam size 4) | 29.16 | --epoch 30, --avg 25, --max-duration 800 |
|
38 |
+
| fast beam search (set as default) | 29.20 | --epoch 30, --avg 25, --max-duration 1500|
|
39 |
+
|
40 |
+
When use-averaged-model=True, the CERs are
|
41 |
+
| | test | comment |
|
42 |
+
|------------------------------------|------------|----------------------------------------------------------------------|
|
43 |
+
| greedy search | 29.89 | --iter 36000, --avg 8, --max-duration 800 --use-averaged-model=True |
|
44 |
+
| modified beam search (beam size 4) | 28.91 | --iter 36000, --avg 8, --max-duration 800 --use-averaged-model=True |
|
45 |
+
| fast beam search (set as default) | 29.08 | --iter 36000, --avg 8, --max-duration 1500 --use-averaged-model=True |
|
46 |
+
|