mirror of
https://github.com/modelscope/FunASR
synced 2025-09-15 14:48:36 +08:00
269c201429
* support resume model from pai * add padding for streaming rnnt conv input * fix large dataset training bug * bug fix * modify aishell rnnt egs to support wav input * add libri_100 rnnt recipe * bug fix * add librispeech rnnt recipe * add librispeech README * update rnnt results * bug fix --------- Co-authored-by: aky15 <ankeyu.aky@11.17.44.249>
19 lines
498 B
Markdown
19 lines
498 B
Markdown
|
|
# Streaming RNN-T Result
|
|
|
|
## Training Config
|
|
- 8 gpu(Tesla V100)
|
|
- Feature info: using 80 dims fbank, global cmvn, speed perturb(0.9, 1.0, 1.1), specaugment
|
|
- Train config: conf/train_conformer_rnnt_unified.yaml
|
|
- chunk config: chunk size 16, 1 left chunk
|
|
- LM config: LM was not used
|
|
- Model size: 90M
|
|
|
|
## Results (CER)
|
|
- Decode config: conf/decode_rnnt_conformer_streaming.yaml
|
|
|
|
| testset | WER(%) |
|
|
|:--------------:|:-------:|
|
|
| test_clean | 3.58 |
|
|
| test_other | 9.27 |
|