dwgnr commited on
Commit
55bdb54
1 Parent(s): 0aed1a2

update readme

Browse files
Files changed (1) hide show
  1. README.md +47 -35
README.md CHANGED
@@ -2,9 +2,10 @@
2
  language:
3
  - en
4
  thumbnail: null
5
- pipeline_tag: automatic-speech-recognition
6
  tags:
 
7
  - CTC
 
8
  - pytorch
9
  - speechbrain
10
  license: apache-2.0
@@ -12,7 +13,7 @@ datasets:
12
  - switchboard
13
  metrics:
14
  - wer
15
- - ser
16
 
17
  ---
18
 
@@ -21,26 +22,23 @@ metrics:
21
 
22
  # wav2vec 2.0 with CTC/Attention trained on Switchboard (No LM)
23
 
24
- This repository provides all the necessary tools to perform automatic speech
25
- recognition from an end-to-end system pretrained on the Switchboard corpus within
26
- SpeechBrain. For a better experience, we encourage you to learn more about
27
- [SpeechBrain](https://speechbrain.github.io).
28
 
29
  The performance of the model is the following:
30
 
31
- | Release | Swbd SER | Callhome SER | Eval2000 SER | Swbd WER | Callhome WER | Eval2000 WER | GPUs |
32
  |:--------:|:--------:|:------------:|:------------:|:--------:|:------------:|:------------:|:-----------:|
33
- | 17-09-22 | 48.60 | 55.76 | 52.96 | 8 .76 | 14.67 | 11.78 | 4xA100 40GB |
34
 
35
- ## Pipeline description
36
 
37
  This ASR system is composed of 2 different but linked blocks:
38
- - Tokenizer (unigram) that transforms words into subword units trained on the Switchboard training transcripts and the Fisher corpus.
39
- - Acoustic model (wav2vec2.0 + CTC). A pretrained wav2vec 2.0 model ([facebook/wav2vec2-large-lv60](https://huggingface.co/facebook/wav2vec2-large-lv60)) is combined with two DNN layers and finetuned on Switchboard
40
- The obtained final acoustic representation is given to the CTC greedy decoder.
41
 
42
  The system is trained with recordings sampled at 16kHz (single channel).
43
- The code will automatically normalize your audio (i.e., resampling + mono channel selection) when calling *transcribe_file* if needed.
44
 
45
  ## Install SpeechBrain
46
 
@@ -53,7 +51,7 @@ pip install speechbrain transformers
53
  Please notice that we encourage you to read our tutorials and learn more about
54
  [SpeechBrain](https://speechbrain.github.io).
55
 
56
- ### Transcribing your own audio files
57
 
58
  ```python
59
  from speechbrain.pretrained import EncoderASR
@@ -62,17 +60,21 @@ asr_model = EncoderASR.from_hparams(source="speechbrain/asr-wav2vec2-switchboard
62
  asr_model.transcribe_file('path/to/audiofile')
63
 
64
  ```
65
- ### Inference on GPU
 
 
66
  To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.
67
 
68
- ### Training
69
- The model was trained with SpeechBrain (Commit hash: '70904d0').
 
70
  To train it from scratch follow these steps:
71
 
72
  1. Clone SpeechBrain:
73
  ```bash
74
  git clone https://github.com/speechbrain/speechbrain/
75
  ```
 
76
  2. Install it:
77
  ```bash
78
  cd speechbrain
@@ -82,32 +84,42 @@ pip install -e .
82
 
83
  3. Run Training:
84
  ```bash
85
- cd recipes/Switchboard/ASR/CTC/
86
  python train_with_wav2vec.py hparams/train_with_wav2vec.yaml --data_folder=your_data_folder
87
  ```
88
 
89
- ### Limitations
90
- The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
91
 
92
- #### Referencing SpeechBrain
93
 
94
- ```
95
- @misc{SB2021,
96
- author = {Ravanelli, Mirco and Parcollet, Titouan and Rouhe, Aku and Plantinga, Peter and Rastorgueva, Elena and Lugosch, Loren and Dawalatabad, Nauman and Ju-Chieh, Chou and Heba, Abdel and Grondin, Francois and Aris, William and Liao, Chien-Feng and Cornell, Samuele and Yeh, Sung-Lin and Na, Hwidong and Gao, Yan and Fu, Szu-Wei and Subakan, Cem and De Mori, Renato and Bengio, Yoshua },
97
- title = {SpeechBrain},
98
- year = {2021},
99
- publisher = {GitHub},
100
- journal = {GitHub repository},
101
- howpublished = {\\\\url{https://github.com/speechbrain/speechbrain}},
102
- }
103
- ```
104
 
105
- #### About SpeechBrain
106
- SpeechBrain is an open-source and all-in-one speech toolkit. It is designed to be simple, extremely flexible, and user-friendly. Competitive or state-of-the-art performance is obtained in various domains.
107
 
108
- Website: https://speechbrain.github.io/
 
 
109
 
110
- GitHub: https://github.com/speechbrain/speechbrain
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111
 
112
 
113
 
 
2
  language:
3
  - en
4
  thumbnail: null
 
5
  tags:
6
+ - automatic-speech-recognition
7
  - CTC
8
+ - Attention
9
  - pytorch
10
  - speechbrain
11
  license: apache-2.0
 
13
  - switchboard
14
  metrics:
15
  - wer
16
+ - cer
17
 
18
  ---
19
 
 
22
 
23
  # wav2vec 2.0 with CTC/Attention trained on Switchboard (No LM)
24
 
25
+ This repository provides all the necessary tools to perform automatic speech recognition from an end-to-end system pretrained on the Switchboard (EN) corpus within SpeechBrain.
26
+ For a better experience, we encourage you to learn more about [SpeechBrain](https://speechbrain.github.io).
 
 
27
 
28
  The performance of the model is the following:
29
 
30
+ | Release | Swbd CER | Callhome CER | Eval2000 CER | Swbd WER | Callhome WER | Eval2000 WER | GPUs |
31
  |:--------:|:--------:|:------------:|:------------:|:--------:|:------------:|:------------:|:-----------:|
32
+ | 17-09-22 | 5.24 | 9.69 | 7.44 | 8 .76 | 14.67 | 11.78 | 4xA100 40GB |
33
 
34
+ ## Pipeline Description
35
 
36
  This ASR system is composed of 2 different but linked blocks:
37
+ - Tokenizer (unigram) that transforms words into subword units trained on the Switchboard training transcriptions and the Fisher corpus.
38
+ - Acoustic model (wav2vec2.0 + CTC). A pretrained wav2vec 2.0 model ([facebook/wav2vec2-large-lv60](https://huggingface.co/facebook/wav2vec2-large-lv60)) is combined with a feature encoder consisting of three DNN layers and finetuned on Switchboard. The obtained final acoustic representation is given to a greedy CTC decoder.
 
39
 
40
  The system is trained with recordings sampled at 16kHz (single channel).
41
+ The code will automatically normalize your audio (i.e., resampling + mono channel selection) when calling `transcribe_file` if needed.
42
 
43
  ## Install SpeechBrain
44
 
 
51
  Please notice that we encourage you to read our tutorials and learn more about
52
  [SpeechBrain](https://speechbrain.github.io).
53
 
54
+ ## Transcribing Your Own Audio Files
55
 
56
  ```python
57
  from speechbrain.pretrained import EncoderASR
 
60
  asr_model.transcribe_file('path/to/audiofile')
61
 
62
  ```
63
+
64
+ ## Inference on GPU
65
+
66
  To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.
67
 
68
+ ## Training
69
+
70
+ The model was trained with SpeechBrain (commit hash: `70904d0`).
71
  To train it from scratch follow these steps:
72
 
73
  1. Clone SpeechBrain:
74
  ```bash
75
  git clone https://github.com/speechbrain/speechbrain/
76
  ```
77
+
78
  2. Install it:
79
  ```bash
80
  cd speechbrain
 
84
 
85
  3. Run Training:
86
  ```bash
87
+ cd recipes/Switchboard/ASR/CTC
88
  python train_with_wav2vec.py hparams/train_with_wav2vec.yaml --data_folder=your_data_folder
89
  ```
90
 
91
+ ## Limitations
 
92
 
93
+ The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
94
 
95
+ ## Credits
96
+
97
+ This model was trained with resources provided by the [THN Center for AI](https://www.th-nuernberg.de/en/kiz).
98
+
99
+ # About SpeechBrain
 
 
 
 
 
100
 
101
+ SpeechBrain is an open-source and all-in-one speech toolkit. It is designed to be simple, extremely flexible, and user-friendly.
102
+ Competitive or state-of-the-art performance is obtained in various domains.
103
 
104
+ - Website: https://speechbrain.github.io/
105
+ - GitHub: https://github.com/speechbrain/speechbrain/
106
+ - HuggingFace: https://huggingface.co/speechbrain/
107
 
108
+ # Citing SpeechBrain
109
+
110
+ Please cite SpeechBrain if you use it for your research or business.
111
+
112
+ ```bibtex
113
+ @misc{speechbrain,
114
+ title={{SpeechBrain}: A General-Purpose Speech Toolkit},
115
+ author={Mirco Ravanelli and Titouan Parcollet and Peter Plantinga and Aku Rouhe and Samuele Cornell and Loren Lugosch and Cem Subakan and Nauman Dawalatabad and Abdelwahab Heba and Jianyuan Zhong and Ju-Chieh Chou and Sung-Lin Yeh and Szu-Wei Fu and Chien-Feng Liao and Elena Rastorgueva and François Grondin and William Aris and Hwidong Na and Yan Gao and Renato De Mori and Yoshua Bengio},
116
+ year={2021},
117
+ eprint={2106.04624},
118
+ archivePrefix={arXiv},
119
+ primaryClass={eess.AS},
120
+ note={arXiv:2106.04624}
121
+ }
122
+ ```
123
 
124
 
125