|
|
|
@ -173,12 +173,7 @@ bash local/data.sh --stage 2 --stop_stage 2
|
|
|
|
|
|
|
|
|
|
CUDA_VISIBLE_DEVICES= ./local/test.sh conf/deepspeech2.yaml exp/deepspeech2/checkpoints/avg_1
|
|
|
|
|
```
|
|
|
|
|
The performance of the released models are shown below:
|
|
|
|
|
|
|
|
|
|
| Acoustic Model | Training Data | Token-based | Size | Descriptions | CER | WER | Hours of speech |
|
|
|
|
|
| :----------------------------: | :-------------: | :---------: | -----: | :------------------------------------------------- | :---- | :--- | :-------------- |
|
|
|
|
|
| Ds2 Online Aishell ASR0 Model | Aishell Dataset | Char-based | 345 MB | 2 Conv + 5 LSTM layers with only forward direction | 0.080 | - | 151 h |
|
|
|
|
|
| Ds2 Offline Aishell ASR0 Model | Aishell Dataset | Char-based | 306 MB | 2 Conv + 3 bidirectional GRU layers | 0.064 | - | 151 h |
|
|
|
|
|
The performance of the released models are shown in [this](./RESULT.md)
|
|
|
|
|
## Stage 4: Static graph model Export
|
|
|
|
|
This stage is to transform dygraph to static graph.
|
|
|
|
|
```bash
|
|
|
|
|