vnoroozi commited on
Commit
982a5ce
1 Parent(s): 7469505

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +175 -0
README.md CHANGED
@@ -1,3 +1,178 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: cc-by-4.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - en
4
+ library_name: nemo
5
+ datasets:
6
+ - librispeech_asr
7
+ - fisher_corpus
8
+ - Switchboard-1
9
+ - WSJ-0
10
+ - WSJ-1
11
+ - National-Singapore-Corpus-Part-1
12
+ - National-Singapore-Corpus-Part-6
13
+ - vctk
14
+ - VoxPopuli-(EN)
15
+ - Europarl-ASR-(EN)
16
+ - Multilingual-LibriSpeech-(2000-hours)
17
+ - mozilla-foundation/common_voice_8_0
18
+ - MLCommons/peoples_speech
19
+ thumbnail: null
20
+ tags:
21
+ - automatic-speech-recognition
22
+ - speech
23
+ - streaming
24
+ - audio
25
+ - Transducer
26
+ - Conformer
27
+ - CTC
28
+ - pytorch
29
+ - NeMo
30
  license: cc-by-4.0
31
+ widget:
32
+ - example_title: Librispeech sample 1
33
+ src: https://cdn-media.huggingface.co/speech_samples/sample1.flac
34
+ - example_title: Librispeech sample 2
35
+ src: https://cdn-media.huggingface.co/speech_samples/sample2.flac
36
+ model-index:
37
+ - name: stt_en_fastconformer_hybrid_large_streaming_multi
38
+ results:
39
+ - task:
40
+ type: Automatic Speech Recognition
41
+ name: automatic-speech-recognition
42
+ dataset:
43
+ name: LibriSpeech (other)
44
+ type: librispeech_asr
45
+ config: other
46
+ split: test
47
+ args:
48
+ language: en
49
+ metrics:
50
+ - name: Test WER
51
+ type: wer
52
+ value: 5.4
53
  ---
54
+
55
+ # NVIDIA Streaming Conformer-Hybrid Large (en-US)
56
+
57
+ <style>
58
+ img {
59
+ display: inline;
60
+ }
61
+ </style>
62
+
63
+ | [![Model architecture](https://img.shields.io/badge/Model_Arch-Conformer--Transducer-lightgrey#model-badge)](#model-architecture)
64
+ | [![Model size](https://img.shields.io/badge/Params-114M-lightgrey#model-badge)](#model-architecture)
65
+ | [![Language](https://img.shields.io/badge/Language-en--US-lightgrey#model-badge)](#datasets)
66
+
67
+
68
+ This model transcribes speech in lower case English alphabet along with spaces and apostrophes.
69
+ It is an "extra-large" versions of Conformer-Transducer (around 600M parameters) model.
70
+ See the [model architecture](#model-architecture) section and [NeMo documentation](https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/models.html#conformer-transducer) for complete architecture details.
71
+
72
+ ## NVIDIA NeMo: Training
73
+
74
+ To train, fine-tune or play with the model you will need to install [NVIDIA NeMo](https://github.com/NVIDIA/NeMo). We recommend you install it after you've installed latest Pytorch version.
75
+ ```
76
+ pip install nemo_toolkit['all']
77
+ '''
78
+ '''
79
+ (if it causes an error):
80
+ pip install nemo_toolkit[all]
81
+ ```
82
+
83
+ ## How to Use this Model
84
+
85
+ The model is available for use in the NeMo toolkit [3], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.
86
+
87
+ ### Automatically instantiate the model
88
+
89
+ ```python
90
+ import nemo.collections.asr as nemo_asr
91
+ asr_model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained("nvidia/stt_en_conformer_transducer_xlarge")
92
+ ```
93
+
94
+ ### Transcribing using Python
95
+ First, let's get a sample
96
+ ```
97
+ wget https://dldata-public.s3.us-east-2.amazonaws.com/2086-149220-0033.wav
98
+ ```
99
+ Then simply do:
100
+ ```
101
+ asr_model.transcribe(['2086-149220-0033.wav'])
102
+ ```
103
+
104
+ ### Transcribing many audio files
105
+
106
+ ```shell
107
+ python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py
108
+ pretrained_name="nvidia/stt_en_conformer_transducer_xlarge"
109
+ audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"
110
+ ```
111
+
112
+ ### Input
113
+
114
+ This model accepts 16000 KHz Mono-channel Audio (wav files) as input.
115
+
116
+ ### Output
117
+
118
+ This model provides transcribed speech as a string for a given audio sample.
119
+
120
+ ## Model Architecture
121
+
122
+ Conformer-Transducer model is an autoregressive variant of Conformer model [1] for Automatic Speech Recognition which uses Transducer loss/decoding instead of CTC Loss. You may find more info on the detail of this model here: [Conformer-Transducer Model](https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/models.html).
123
+
124
+ ## Training
125
+
126
+ The NeMo toolkit [3] was used for training the models for over several hundred epochs. These model are trained with this [example script](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/asr_transducer/speech_to_text_rnnt_bpe.py) and this [base config](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/conf/conformer/conformer_transducer_bpe.yaml).
127
+
128
+ The tokenizers for these models were built using the text transcripts of the train set with this [script](https://github.com/NVIDIA/NeMo/blob/main/scripts/tokenizers/process_asr_text_tokenizer.py).
129
+
130
+ ### Datasets
131
+
132
+ All the models in this collection are trained on a composite dataset (NeMo ASRSET) comprising of several thousand hours of English speech:
133
+
134
+ - Librispeech 960 hours of English speech
135
+ - Fisher Corpus
136
+ - Switchboard-1 Dataset
137
+ - WSJ-0 and WSJ-1
138
+ - National Speech Corpus (Part 1, Part 6)
139
+ - VCTK
140
+ - VoxPopuli (EN)
141
+ - Europarl-ASR (EN)
142
+ - Multilingual Librispeech (MLS EN) - 2,000 hrs subset
143
+ - Mozilla Common Voice (v8.0)
144
+ - People's Speech - 12,000 hrs subset
145
+
146
+ Note: older versions of the model may have trained on smaller set of datasets.
147
+
148
+ ## Performance
149
+
150
+ The list of the available models in this collection is shown in the following table. Performances of the ASR models are reported in terms of Word Error Rate (WER%) with greedy decoding.
151
+
152
+ | Version | Tokenizer | Vocabulary Size | LS test-other | LS test-clean | WSJ Eval92 | WSJ Dev93 | NSC Part 1 | MLS Test | MLS Dev | MCV Test 8.0 | Train Dataset |
153
+ |---------|-----------------------|-----------------|---------------|---------------|------------|-----------|-----|-------|------|----|------|
154
+ | 1.10.0 | SentencePiece Unigram | 1024 | 3.01 | 1.62 | 1.17 | 2.05 | 5.70 | 5.32 | 4.59 | 6.46 | NeMo ASRSET 3.0 |
155
+
156
+ ## Limitations
157
+ Since this model was trained on publicly available speech datasets, the performance of this model might degrade for speech which includes technical terms, or vernacular that the model has not been trained on. The model might also perform worse for accented speech.
158
+
159
+ ## NVIDIA Riva: Deployment
160
+
161
+ [NVIDIA Riva](https://developer.nvidia.com/riva), is an accelerated speech AI SDK deployable on-prem, in all clouds, multi-cloud, hybrid, on edge, and embedded.
162
+ Additionally, Riva provides:
163
+
164
+ * World-class out-of-the-box accuracy for the most common languages with model checkpoints trained on proprietary data with hundreds of thousands of GPU-compute hours
165
+ * Best in class accuracy with run-time word boosting (e.g., brand and product names) and customization of acoustic model, language model, and inverse text normalization
166
+ * Streaming speech recognition, Kubernetes compatible scaling, and enterprise-grade support
167
+
168
+ Although this model isn’t supported yet by Riva, the [list of supported models is here](https://huggingface.co/models?other=Riva).
169
+ Check out [Riva live demo](https://developer.nvidia.com/riva#demos).
170
+
171
+ ## References
172
+ [1] [Conformer: Convolution-augmented Transformer for Speech Recognition](https://arxiv.org/abs/2005.08100)
173
+ [2] [Google Sentencepiece Tokenizer](https://github.com/google/sentencepiece)
174
+ [3] [NVIDIA NeMo Toolkit](https://github.com/NVIDIA/NeMo)
175
+
176
+ ## Licence
177
+
178
+ License to use this model is covered by the [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/). By downloading the public and release version of the model, you accept the terms and conditions of the [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) license.