Andreas Nautsch
commited on
Commit
•
8a0c123
1
Parent(s):
637a618
Update README.md
Browse files
README.md
CHANGED
@@ -14,71 +14,102 @@ metrics:
|
|
14 |
- wer
|
15 |
- cer
|
16 |
---
|
|
|
17 |
<iframe src="https://ghbtns.com/github-btn.html?user=speechbrain&repo=speechbrain&type=star&count=true&size=large&v=2" frameborder="0" scrolling="0" width="170" height="30" title="GitHub"></iframe>
|
18 |
<br/><br/>
|
|
|
19 |
# CRDNN with CTC/Attention trained on CommonVoice 7.0 German (No LM)
|
20 |
This repository provides all the necessary tools to perform automatic speech
|
21 |
recognition from an end-to-end system pretrained on CommonVoice (German Language) within
|
22 |
SpeechBrain. For a better experience, we encourage you to learn more about
|
23 |
[SpeechBrain](https://speechbrain.github.io).
|
24 |
The performance of the model is the following:
|
|
|
25 |
| Release | Test CER | Test WER | GPUs |
|
26 |
|:-------------:|:--------------:|:--------------:| :--------:|
|
27 |
-
| 28
|
|
|
28 |
## Credits
|
29 |
The model is provided by [vitas.ai](vitas.ai).
|
|
|
30 |
## Pipeline description
|
31 |
This ASR system is composed of 2 different but linked blocks:
|
|
|
32 |
- Tokenizer (unigram) that transforms words into subword units and trained with
|
33 |
the train transcriptions (train.tsv) of CommonVoice (DE).
|
34 |
- Acoustic model (CRDNN + CTC/Attention). The CRDNN architecture is made of
|
35 |
N blocks of convolutional neural networks with normalization and pooling on the
|
36 |
frequency domain. Then, a bidirectional LSTM is connected to a final DNN to obtain
|
37 |
the final acoustic representation that is given to the CTC and attention decoders.
|
|
|
38 |
## Install SpeechBrain
|
39 |
First of all, please install SpeechBrain with the following command:
|
|
|
40 |
```
|
41 |
pip install speechbrain
|
42 |
```
|
|
|
43 |
Please notice that we encourage you to read our tutorials and learn more about
|
44 |
[SpeechBrain](https://speechbrain.github.io).
|
|
|
45 |
### Transcribing your own audio files (in German)
|
|
|
46 |
```python
|
47 |
from speechbrain.pretrained import EncoderDecoderASR
|
48 |
asr_model = EncoderDecoderASR.from_hparams(source="speechbrain/asr-crdnn-commonvoice-de", savedir="pretrained_models/asr-crdnn-commonvoice-de")
|
49 |
asr_model.transcribe_file("speechbrain/asr-crdnn-commonvoice-de/example-de.wav")
|
50 |
```
|
|
|
51 |
### Inference on GPU
|
|
|
52 |
To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.
|
|
|
53 |
## Parallel Inference on a Batch
|
|
|
54 |
Please, [see this Colab notebook](https://colab.research.google.com/drive/1hX5ZI9S4jHIjahFCZnhwwQmFoGAi3tmu?usp=sharing) to figure out how to transcribe in parallel a batch of input sentences using a pre-trained model.
|
|
|
55 |
### Training
|
|
|
56 |
The model was trained with SpeechBrain (986a2175).
|
57 |
To train it from scratch follows these steps:
|
|
|
58 |
1. Clone SpeechBrain:
|
|
|
59 |
```bash
|
60 |
git clone https://github.com/speechbrain/speechbrain/
|
61 |
```
|
|
|
62 |
2. Install it:
|
|
|
63 |
```
|
64 |
cd speechbrain
|
65 |
pip install -r requirements.txt
|
66 |
pip install -e .
|
67 |
```
|
|
|
68 |
3. Run Training:
|
|
|
69 |
```
|
70 |
cd recipes/CommonVoice/ASR/seq2seq
|
71 |
python train.py hparams/train_de.yaml --data_folder=your_data_folder
|
72 |
```
|
|
|
73 |
You can find our training results (models, logs, etc) [here](https://drive.google.com/drive/folders/13i7rdgVX7-qZ94Rtj6OdUgU-S6BbKKvw?usp=sharing)
|
|
|
74 |
### Limitations
|
|
|
75 |
The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
|
|
|
76 |
# **About SpeechBrain**
|
|
|
77 |
- Website: https://speechbrain.github.io/
|
78 |
- Code: https://github.com/speechbrain/speechbrain/
|
79 |
- HuggingFace: https://huggingface.co/speechbrain/
|
|
|
80 |
# **Citing SpeechBrain**
|
|
|
81 |
Please, cite SpeechBrain if you use it for your research or business.
|
|
|
82 |
```bibtex
|
83 |
@misc{speechbrain,
|
84 |
title={{SpeechBrain}: A General-Purpose Speech Toolkit},
|
|
|
14 |
- wer
|
15 |
- cer
|
16 |
---
|
17 |
+
|
18 |
<iframe src="https://ghbtns.com/github-btn.html?user=speechbrain&repo=speechbrain&type=star&count=true&size=large&v=2" frameborder="0" scrolling="0" width="170" height="30" title="GitHub"></iframe>
|
19 |
<br/><br/>
|
20 |
+
|
21 |
# CRDNN with CTC/Attention trained on CommonVoice 7.0 German (No LM)
|
22 |
This repository provides all the necessary tools to perform automatic speech
|
23 |
recognition from an end-to-end system pretrained on CommonVoice (German Language) within
|
24 |
SpeechBrain. For a better experience, we encourage you to learn more about
|
25 |
[SpeechBrain](https://speechbrain.github.io).
|
26 |
The performance of the model is the following:
|
27 |
+
|
28 |
| Release | Test CER | Test WER | GPUs |
|
29 |
|:-------------:|:--------------:|:--------------:| :--------:|
|
30 |
+
| 28.10.21 | 4.93 | 15.37 | 1xV100 16GB |
|
31 |
+
|
32 |
## Credits
|
33 |
The model is provided by [vitas.ai](vitas.ai).
|
34 |
+
|
35 |
## Pipeline description
|
36 |
This ASR system is composed of 2 different but linked blocks:
|
37 |
+
|
38 |
- Tokenizer (unigram) that transforms words into subword units and trained with
|
39 |
the train transcriptions (train.tsv) of CommonVoice (DE).
|
40 |
- Acoustic model (CRDNN + CTC/Attention). The CRDNN architecture is made of
|
41 |
N blocks of convolutional neural networks with normalization and pooling on the
|
42 |
frequency domain. Then, a bidirectional LSTM is connected to a final DNN to obtain
|
43 |
the final acoustic representation that is given to the CTC and attention decoders.
|
44 |
+
|
45 |
## Install SpeechBrain
|
46 |
First of all, please install SpeechBrain with the following command:
|
47 |
+
|
48 |
```
|
49 |
pip install speechbrain
|
50 |
```
|
51 |
+
|
52 |
Please notice that we encourage you to read our tutorials and learn more about
|
53 |
[SpeechBrain](https://speechbrain.github.io).
|
54 |
+
|
55 |
### Transcribing your own audio files (in German)
|
56 |
+
|
57 |
```python
|
58 |
from speechbrain.pretrained import EncoderDecoderASR
|
59 |
asr_model = EncoderDecoderASR.from_hparams(source="speechbrain/asr-crdnn-commonvoice-de", savedir="pretrained_models/asr-crdnn-commonvoice-de")
|
60 |
asr_model.transcribe_file("speechbrain/asr-crdnn-commonvoice-de/example-de.wav")
|
61 |
```
|
62 |
+
|
63 |
### Inference on GPU
|
64 |
+
|
65 |
To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.
|
66 |
+
|
67 |
## Parallel Inference on a Batch
|
68 |
+
|
69 |
Please, [see this Colab notebook](https://colab.research.google.com/drive/1hX5ZI9S4jHIjahFCZnhwwQmFoGAi3tmu?usp=sharing) to figure out how to transcribe in parallel a batch of input sentences using a pre-trained model.
|
70 |
+
|
71 |
### Training
|
72 |
+
|
73 |
The model was trained with SpeechBrain (986a2175).
|
74 |
To train it from scratch follows these steps:
|
75 |
+
|
76 |
1. Clone SpeechBrain:
|
77 |
+
|
78 |
```bash
|
79 |
git clone https://github.com/speechbrain/speechbrain/
|
80 |
```
|
81 |
+
|
82 |
2. Install it:
|
83 |
+
|
84 |
```
|
85 |
cd speechbrain
|
86 |
pip install -r requirements.txt
|
87 |
pip install -e .
|
88 |
```
|
89 |
+
|
90 |
3. Run Training:
|
91 |
+
|
92 |
```
|
93 |
cd recipes/CommonVoice/ASR/seq2seq
|
94 |
python train.py hparams/train_de.yaml --data_folder=your_data_folder
|
95 |
```
|
96 |
+
|
97 |
You can find our training results (models, logs, etc) [here](https://drive.google.com/drive/folders/13i7rdgVX7-qZ94Rtj6OdUgU-S6BbKKvw?usp=sharing)
|
98 |
+
|
99 |
### Limitations
|
100 |
+
|
101 |
The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
|
102 |
+
|
103 |
# **About SpeechBrain**
|
104 |
+
|
105 |
- Website: https://speechbrain.github.io/
|
106 |
- Code: https://github.com/speechbrain/speechbrain/
|
107 |
- HuggingFace: https://huggingface.co/speechbrain/
|
108 |
+
|
109 |
# **Citing SpeechBrain**
|
110 |
+
|
111 |
Please, cite SpeechBrain if you use it for your research or business.
|
112 |
+
|
113 |
```bibtex
|
114 |
@misc{speechbrain,
|
115 |
title={{SpeechBrain}: A General-Purpose Speech Toolkit},
|