poonehmousavi commited on
Commit
418d5a5
1 Parent(s): 50ebe46

Upload 5 files

Browse files
Files changed (5) hide show
  1. README.md +114 -1
  2. config.json +69 -0
  3. example.wav +0 -0
  4. hyperparams.yaml +118 -0
  5. preprocessor_config.json +8 -0
README.md CHANGED
@@ -1,3 +1,116 @@
1
  ---
2
- license: apache-2.0
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language: "en"
3
+ thumbnail:
4
+ pipeline_tag: automatic-speech-recognition
5
+ tags:
6
+ - CTC
7
+ - pytorch
8
+ - speechbrain
9
+ - Transformer
10
+ license: "apache-2.0"
11
+ datasets:
12
+ - commonvoice
13
+ metrics:
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
+ # wav2vec 2.0 with CTC trained on CommonVoice English (No LM)
22
+
23
+ This repository provides all the necessary tools to perform automatic speech
24
+ recognition from an end-to-end system pretrained on CommonVoice (English Language) within
25
+ SpeechBrain. For a better experience, we encourage you to learn more about
26
+ [SpeechBrain](https://speechbrain.github.io).
27
+
28
+ The performance of the model is the following:
29
+
30
+ | Release | Test WER | GPUs |
31
+ |:--------------:|:--------------:| :--------:|
32
+ | 03-06-21 | 15.69 | 2xV100 32GB |
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 and trained with
38
+ the train transcriptions (train.tsv) of CommonVoice (EN).
39
+ - Acoustic model (wav2vec2.0 + CTC). A pretrained wav2vec 2.0 model ([wav2vec2-lv60-large](https://huggingface.co/facebook/wav2vec2-large-lv60)) is combined with two DNN layers and finetuned on CommonVoice En.
40
+ The obtained final acoustic representation is given to the CTC 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
+
47
+ First of all, please install tranformers and SpeechBrain with the following command:
48
+
49
+ ```
50
+ pip install speechbrain transformers
51
+ ```
52
+
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 (in English)
57
+
58
+ ```python
59
+ from speechbrain.pretrained import EncoderDecoderASR
60
+
61
+ asr_model = EncoderDecoderASR.from_hparams(source="speechbrain/asr-wav2vec2-commonvoice-en", savedir="pretrained_models/asr-wav2vec2-commonvoice-en")
62
+ asr_model.transcribe_file("speechbrain/asr-wav2vec2-commonvoice-en/example.wav")
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
+ ## Parallel Inference on a Batch
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
+ The model was trained with SpeechBrain.
73
+ To train it from scratch follow these steps:
74
+ 1. Clone SpeechBrain:
75
+ ```bash
76
+ git clone https://github.com/speechbrain/speechbrain/
77
+ ```
78
+ 2. Install it:
79
+ ```bash
80
+ cd speechbrain
81
+ pip install -r requirements.txt
82
+ pip install -e .
83
+ ```
84
+
85
+ 3. Run Training:
86
+ ```bash
87
+ cd recipes/CommonVoice/ASR/seq2seq
88
+ python train.py hparams/train_en_with_wav2vec.yaml --data_folder=your_data_folder
89
+ ```
90
+
91
+ You can find our training results (models, logs, etc) [here](https://drive.google.com/drive/folders/1tjz6IZmVRkuRE97E7h1cXFoGTer7pT73?usp=sharing).
92
+
93
+ ### Limitations
94
+ The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
95
+
96
+
97
+ # **About SpeechBrain**
98
+ - Website: https://speechbrain.github.io/
99
+ - Code: https://github.com/speechbrain/speechbrain/
100
+ - HuggingFace: https://huggingface.co/speechbrain/
101
+
102
+
103
+ # **Citing SpeechBrain**
104
+ Please, cite SpeechBrain if you use it for your research or business.
105
+
106
+ ```bibtex
107
+ @misc{speechbrain,
108
+ title={{SpeechBrain}: A General-Purpose Speech Toolkit},
109
+ 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},
110
+ year={2021},
111
+ eprint={2106.04624},
112
+ archivePrefix={arXiv},
113
+ primaryClass={eess.AS},
114
+ note={arXiv:2106.04624}
115
+ }
116
+ ```
config.json ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "speechbrain_interface": "EncoderDecoderASR",
3
+ "activation_dropout": 0.1,
4
+ "apply_spec_augment": true,
5
+ "architectures": [
6
+ "Wav2Vec2Model"
7
+ ],
8
+ "attention_dropout": 0.1,
9
+ "bos_token_id": 1,
10
+ "conv_bias": true,
11
+ "conv_dim": [
12
+ 512,
13
+ 512,
14
+ 512,
15
+ 512,
16
+ 512,
17
+ 512,
18
+ 512
19
+ ],
20
+ "conv_kernel": [
21
+ 10,
22
+ 3,
23
+ 3,
24
+ 3,
25
+ 3,
26
+ 2,
27
+ 2
28
+ ],
29
+ "conv_stride": [
30
+ 5,
31
+ 2,
32
+ 2,
33
+ 2,
34
+ 2,
35
+ 2,
36
+ 2
37
+ ],
38
+ "ctc_loss_reduction": "sum",
39
+ "ctc_zero_infinity": false,
40
+ "do_stable_layer_norm": true,
41
+ "eos_token_id": 2,
42
+ "feat_extract_activation": "gelu",
43
+ "feat_extract_dropout": 0.0,
44
+ "feat_extract_norm": "layer",
45
+ "feat_proj_dropout": 0.1,
46
+ "final_dropout": 0.1,
47
+ "gradient_checkpointing": false,
48
+ "hidden_act": "gelu",
49
+ "hidden_dropout": 0.1,
50
+ "hidden_dropout_prob": 0.1,
51
+ "hidden_size": 1024,
52
+ "initializer_range": 0.02,
53
+ "intermediate_size": 4096,
54
+ "layer_norm_eps": 1e-05,
55
+ "layerdrop": 0.1,
56
+ "mask_feature_length": 10,
57
+ "mask_feature_prob": 0.0,
58
+ "mask_time_length": 10,
59
+ "mask_time_prob": 0.05,
60
+ "model_type": "wav2vec2",
61
+ "num_attention_heads": 16,
62
+ "num_conv_pos_embedding_groups": 16,
63
+ "num_conv_pos_embeddings": 128,
64
+ "num_feat_extract_layers": 7,
65
+ "num_hidden_layers": 24,
66
+ "pad_token_id": 0,
67
+ "transformers_version": "4.4.0.dev0",
68
+ "vocab_size": 32
69
+ }
example.wav ADDED
Binary file (104 kB). View file
 
hyperparams.yaml ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # ################################
2
+ # Model: wav2vec2 + DNN + CTC/Attention
3
+ # Augmentation: SpecAugment
4
+ # Authors: Titouan Parcollet 2021
5
+ # ################################
6
+
7
+ sample_rate: 16000
8
+ wav2vec2_hub: facebook/wav2vec2-large-lv60
9
+
10
+ # BPE parameters
11
+ token_type: unigram # ["unigram", "bpe", "char"]
12
+ character_coverage: 1.0
13
+
14
+ # Model parameters
15
+ activation: !name:torch.nn.LeakyReLU
16
+ dnn_layers: 2
17
+ dnn_neurons: 1024
18
+ emb_size: 128
19
+ dec_neurons: 1024
20
+
21
+ # Outputs
22
+ output_neurons: 1000 # BPE size, index(blank/eos/bos) = 0
23
+
24
+ # Decoding parameters
25
+ # Be sure that the bos and eos index match with the BPEs ones
26
+ blank_index: 0
27
+ bos_index: 1
28
+ eos_index: 2
29
+ min_decode_ratio: 0.0
30
+ max_decode_ratio: 1.0
31
+ beam_size: 10
32
+ eos_threshold: 1.5
33
+ using_max_attn_shift: True
34
+ max_attn_shift: 140
35
+ ctc_weight_decode: 0.0
36
+ temperature: 1.50
37
+
38
+ enc: !new:speechbrain.lobes.models.VanillaNN.VanillaNN
39
+ input_shape: [null, null, 1024]
40
+ activation: !ref <activation>
41
+ dnn_blocks: !ref <dnn_layers>
42
+ dnn_neurons: !ref <dnn_neurons>
43
+
44
+ wav2vec2: !new:speechbrain.lobes.models.huggingface_wav2vec.HuggingFaceWav2Vec2
45
+ source: !ref <wav2vec2_hub>
46
+ output_norm: True
47
+ freeze: True
48
+ save_path: model_checkpoints
49
+
50
+ emb: !new:speechbrain.nnet.embedding.Embedding
51
+ num_embeddings: !ref <output_neurons>
52
+ embedding_dim: !ref <emb_size>
53
+
54
+ dec: !new:speechbrain.nnet.RNN.AttentionalRNNDecoder
55
+ enc_dim: !ref <dnn_neurons>
56
+ input_size: !ref <emb_size>
57
+ rnn_type: gru
58
+ attn_type: location
59
+ hidden_size: 1024
60
+ attn_dim: 1024
61
+ num_layers: 1
62
+ scaling: 1.0
63
+ channels: 10
64
+ kernel_size: 100
65
+ re_init: True
66
+ dropout: 0.0
67
+
68
+ ctc_lin: !new:speechbrain.nnet.linear.Linear
69
+ input_size: !ref <dnn_neurons>
70
+ n_neurons: !ref <output_neurons>
71
+
72
+ seq_lin: !new:speechbrain.nnet.linear.Linear
73
+ input_size: !ref <dec_neurons>
74
+ n_neurons: !ref <output_neurons>
75
+
76
+ log_softmax: !new:speechbrain.nnet.activations.Softmax
77
+ apply_log: True
78
+
79
+ ctc_cost: !name:speechbrain.nnet.losses.ctc_loss
80
+ blank_index: !ref <blank_index>
81
+
82
+ seq_cost: !name:speechbrain.nnet.losses.nll_loss
83
+ label_smoothing: 0.1
84
+
85
+ asr_model: !new:torch.nn.ModuleList
86
+ - [!ref <enc>, !ref <emb>, !ref <dec>, !ref <ctc_lin>, !ref <seq_lin>]
87
+
88
+ tokenizer: !new:sentencepiece.SentencePieceProcessor
89
+
90
+ encoder: !new:speechbrain.nnet.containers.LengthsCapableSequential
91
+ wav2vec2: !ref <wav2vec2>
92
+ enc: !ref <enc>
93
+
94
+ decoder: !new:speechbrain.decoders.S2SRNNBeamSearcher
95
+ embedding: !ref <emb>
96
+ decoder: !ref <dec>
97
+ linear: !ref <seq_lin>
98
+ ctc_linear: !ref <ctc_lin>
99
+ bos_index: !ref <bos_index>
100
+ eos_index: !ref <eos_index>
101
+ blank_index: !ref <blank_index>
102
+ min_decode_ratio: !ref <min_decode_ratio>
103
+ max_decode_ratio: !ref <max_decode_ratio>
104
+ beam_size: !ref <beam_size>
105
+ eos_threshold: !ref <eos_threshold>
106
+ using_max_attn_shift: !ref <using_max_attn_shift>
107
+ max_attn_shift: !ref <max_attn_shift>
108
+ temperature: !ref <temperature>
109
+
110
+ modules:
111
+ encoder: !ref <encoder>
112
+ decoder: !ref <decoder>
113
+
114
+ pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
115
+ loadables:
116
+ wav2vec2: !ref <wav2vec2>
117
+ asr: !ref <asr_model>
118
+ tokenizer: !ref <tokenizer>
preprocessor_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_normalize": true,
3
+ "feature_size": 1,
4
+ "padding_side": "right",
5
+ "padding_value": 0.0,
6
+ "return_attention_mask": true,
7
+ "sampling_rate": 16000
8
+ }