Commit
•
0361691
1
Parent(s):
bb52bb9
Update README.md
Browse files
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
language:
|
3 |
-
-
|
4 |
thumbnail: null
|
5 |
pipeline_tag: automatic-speech-recognition
|
6 |
tags:
|
@@ -15,31 +15,31 @@ metrics:
|
|
15 |
- wer
|
16 |
- cer
|
17 |
model-index:
|
18 |
-
- name: asr-wav2vec2-commonvoice-14-
|
19 |
results:
|
20 |
- task:
|
21 |
name: Automatic Speech Recognition
|
22 |
type: automatic-speech-recognition
|
23 |
dataset:
|
24 |
-
name: CommonVoice Corpus 14.0 (
|
25 |
type: mozilla-foundation/common_voice_14.0
|
26 |
-
config:
|
27 |
split: test
|
28 |
args:
|
29 |
-
language:
|
30 |
metrics:
|
31 |
- name: Test WER
|
32 |
type: wer
|
33 |
-
value: '
|
34 |
---
|
35 |
|
36 |
<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>
|
37 |
<br/><br/>
|
38 |
|
39 |
-
# wav2vec 2.0 with CTC trained on CommonVoice
|
40 |
|
41 |
This repository provides all the necessary tools to perform automatic speech
|
42 |
-
recognition from an end-to-end system pretrained on CommonVoice (
|
43 |
SpeechBrain. For a better experience, we encourage you to learn more about
|
44 |
[SpeechBrain](https://speechbrain.github.io).
|
45 |
|
@@ -47,14 +47,14 @@ The performance of the model is the following:
|
|
47 |
|
48 |
| Release | Test CER | Test WER | GPUs |
|
49 |
|:-------------:|:--------------:|:--------------:| :--------:|
|
50 |
-
| 15-08-23 |
|
51 |
|
52 |
## Pipeline description
|
53 |
|
54 |
This ASR system is composed of 2 different but linked blocks:
|
55 |
- Tokenizer (unigram) that transforms words into unigrams and trained with
|
56 |
-
the train transcriptions (train.tsv) of CommonVoice (
|
57 |
-
- Acoustic model (wav2vec2.0 + CTC). A pretrained wav2vec 2.0 model ([wav2vec2-large-
|
58 |
The obtained final acoustic representation is given to the CTC decoder.
|
59 |
|
60 |
The system is trained with recordings sampled at 16kHz (single channel).
|
@@ -71,13 +71,13 @@ pip install speechbrain transformers
|
|
71 |
Please notice that we encourage you to read our tutorials and learn more about
|
72 |
[SpeechBrain](https://speechbrain.github.io).
|
73 |
|
74 |
-
### Transcribing your own audio files (in
|
75 |
|
76 |
```python
|
77 |
from speechbrain.pretrained import EncoderASR
|
78 |
|
79 |
-
asr_model = EncoderASR.from_hparams(source="speechbrain/asr-wav2vec2-commonvoice-14-
|
80 |
-
asr_model.transcribe_file("speechbrain/asr-wav2vec2-commonvoice-14-
|
81 |
|
82 |
```
|
83 |
### Inference on GPU
|
@@ -103,10 +103,10 @@ pip install -e .
|
|
103 |
3. Run Training:
|
104 |
```bash
|
105 |
cd recipes/CommonVoice/ASR/CTC/
|
106 |
-
python train_with_wav2vec.py hparams/
|
107 |
```
|
108 |
|
109 |
-
You can find our training results (models, logs, etc) [here](https://www.dropbox.com/sh/
|
110 |
|
111 |
### Limitations
|
112 |
The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
|
|
|
1 |
---
|
2 |
language:
|
3 |
+
- it
|
4 |
thumbnail: null
|
5 |
pipeline_tag: automatic-speech-recognition
|
6 |
tags:
|
|
|
15 |
- wer
|
16 |
- cer
|
17 |
model-index:
|
18 |
+
- name: asr-wav2vec2-commonvoice-14-it
|
19 |
results:
|
20 |
- task:
|
21 |
name: Automatic Speech Recognition
|
22 |
type: automatic-speech-recognition
|
23 |
dataset:
|
24 |
+
name: CommonVoice Corpus 14.0 (Italian)
|
25 |
type: mozilla-foundation/common_voice_14.0
|
26 |
+
config: it
|
27 |
split: test
|
28 |
args:
|
29 |
+
language: it
|
30 |
metrics:
|
31 |
- name: Test WER
|
32 |
type: wer
|
33 |
+
value: '8.28'
|
34 |
---
|
35 |
|
36 |
<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>
|
37 |
<br/><br/>
|
38 |
|
39 |
+
# wav2vec 2.0 with CTC trained on CommonVoice Italian (No LM)
|
40 |
|
41 |
This repository provides all the necessary tools to perform automatic speech
|
42 |
+
recognition from an end-to-end system pretrained on CommonVoice (Italian Language) within
|
43 |
SpeechBrain. For a better experience, we encourage you to learn more about
|
44 |
[SpeechBrain](https://speechbrain.github.io).
|
45 |
|
|
|
47 |
|
48 |
| Release | Test CER | Test WER | GPUs |
|
49 |
|:-------------:|:--------------:|:--------------:| :--------:|
|
50 |
+
| 15-08-23 | 2.38 | 8.38 | 1xV100 32GB |
|
51 |
|
52 |
## Pipeline description
|
53 |
|
54 |
This ASR system is composed of 2 different but linked blocks:
|
55 |
- Tokenizer (unigram) that transforms words into unigrams and trained with
|
56 |
+
the train transcriptions (train.tsv) of CommonVoice (it).
|
57 |
+
- Acoustic model (wav2vec2.0 + CTC). A pretrained wav2vec 2.0 model ([wav2vec2-large-it-voxpopuli](https://huggingface.co/facebook/wav2vec2-large-it-voxpopuli)) is combined with two DNN layers and finetuned on CommonVoice DE.
|
58 |
The obtained final acoustic representation is given to the CTC decoder.
|
59 |
|
60 |
The system is trained with recordings sampled at 16kHz (single channel).
|
|
|
71 |
Please notice that we encourage you to read our tutorials and learn more about
|
72 |
[SpeechBrain](https://speechbrain.github.io).
|
73 |
|
74 |
+
### Transcribing your own audio files (in Italian)
|
75 |
|
76 |
```python
|
77 |
from speechbrain.pretrained import EncoderASR
|
78 |
|
79 |
+
asr_model = EncoderASR.from_hparams(source="speechbrain/asr-wav2vec2-commonvoice-14-it", savedir="pretrained_models/asr-wav2vec2-commonvoice-14-it")
|
80 |
+
asr_model.transcribe_file("speechbrain/asr-wav2vec2-commonvoice-14-it/example-it.wav")
|
81 |
|
82 |
```
|
83 |
### Inference on GPU
|
|
|
103 |
3. Run Training:
|
104 |
```bash
|
105 |
cd recipes/CommonVoice/ASR/CTC/
|
106 |
+
python train_with_wav2vec.py hparams/train_it_with_wav2vec.yaml --data_folder=your_data_folder
|
107 |
```
|
108 |
|
109 |
+
You can find our training results (models, logs, etc) [here](https://www.dropbox.com/sh/hthxqzh5boq15rn/AACftSab_FM6EFWWPgHpKw82a?dl=0).
|
110 |
|
111 |
### Limitations
|
112 |
The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
|