vasista22 commited on
Commit
3a87303
1 Parent(s): b2d33dc

inference_code_snippet_added

Browse files
Files changed (1) hide show
  1. README.md +55 -6
README.md CHANGED
@@ -43,15 +43,62 @@ should probably proofread and complete it, then remove this comment. -->
43
  This model is a fine-tuned version of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2) on the Tamil data available from multiple publicly available ASR corpuses.
44
  It has been fine-tuned as a part of the Whisper fine-tuning sprint.
45
 
46
- ## Training and evaluation data at Speech Lab, IITM
47
 
48
- Training Data: MILE ASR Corpus, ULCA ASR Corpus, Shrutilipi ASR Corpus, Microsoft Research Tamil Corpus (Train+Dev), Babel ASR Corpus, Google/Fleurs (Train+Dev) set.
49
 
50
- Evaluation Data: MILE ASR Corpus Test, Babel Test, Microsoft Research Tamil Corpus Test, Google/Fleurs Test set.
51
 
52
- ## Training procedure
53
 
54
- ### Training hyperparameters
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55
 
56
  The following hyperparameters were used during training:
57
  - learning_rate: 0.75e-05
@@ -65,4 +112,6 @@ The following hyperparameters were used during training:
65
  - mixed_precision_training: True
66
 
67
  ## Acknowledgement
68
- This work was done at Speech Lab, IITM. The compute resources for this work were funded by "Bhashini: National Language translation Mission" project of the Ministry of Electronics and Information Technology (MeitY), Government of India.
 
 
 
43
  This model is a fine-tuned version of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2) on the Tamil data available from multiple publicly available ASR corpuses.
44
  It has been fine-tuned as a part of the Whisper fine-tuning sprint.
45
 
46
+ **NOTE:** The code used to train this model is available for re-use in the [whisper-finetune](https://github.com/vasistalodagala/whisper-finetune) repository.
47
 
48
+ ## Usage
49
 
50
+ In order to evaluate this model on an entire dataset, the evaluation codes available in the [whisper-finetune](https://github.com/vasistalodagala/whisper-finetune) repository can be used.
51
 
52
+ The same repository also provides the scripts for faster inference using whisper-jax.
53
 
54
+ In order to infer a single audio file using this model, the following code snippet can be used:
55
+
56
+ ```python
57
+ >>> import torch
58
+ >>> from transformers import pipeline
59
+
60
+ >>> # path to the audio file to be transcribed
61
+ >>> audio = "/path/to/audio.format"
62
+ >>> device = "cuda:0" if torch.cuda.is_available() else "cpu"
63
+
64
+ >>> transcribe = pipeline(task="automatic-speech-recognition", model="vasista22/whisper-tamil-large-v2", chunk_length_s=30, device=device)
65
+ >>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ta", task="transcribe")
66
+
67
+ >>> print('Transcription: ', transcribe(audio)["text"])
68
+ ```
69
+
70
+ For faster inference of whisper models, the [whisper-jax](https://github.com/sanchit-gandhi/whisper-jax) library can be used. Please follow the necessary installation steps as mentioned [here](https://github.com/vasistalodagala/whisper-finetune#faster-evaluation-with-whisper-jax), before using the following code snippet:
71
+
72
+ ```python
73
+ >>> import jax.numpy as jnp
74
+ >>> from whisper_jax import FlaxWhisperForConditionalGeneration, FlaxWhisperPipline
75
+
76
+ >>> # path to the audio file to be transcribed
77
+ >>> audio = "/path/to/audio.format"
78
+
79
+ >>> transcribe = FlaxWhisperPipline("vasista22/whisper-tamil-large-v2", batch_size=16)
80
+ >>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ta", task="transcribe")
81
+
82
+ >>> print('Transcription: ', transcribe(audio)["text"])
83
+ ```
84
+
85
+ ## Training and evaluation data
86
+
87
+ Training Data:
88
+ - [IISc-MILE Tamil ASR Corpus](https://www.openslr.org/127/)
89
+ - [ULCA ASR Corpus](https://github.com/Open-Speech-EkStep/ULCA-asr-dataset-corpus#tamil-labelled--total-duration-is-116024-hours)
90
+ - [Shrutilipi ASR Corpus](https://ai4bharat.org/shrutilipi)
91
+ - [Microsoft Speech Corpus (Indian Languages)](https://msropendata.com/datasets/7230b4b1-912d-400e-be58-f84e0512985e)
92
+ - [Google/Fleurs Train+Dev set](https://huggingface.co/datasets/google/fleurs)
93
+ - Babel ASR Corpus
94
+
95
+ Evaluation Data:
96
+ - [Microsoft Speech Corpus (Indian Languages) Test Set](https://msropendata.com/datasets/7230b4b1-912d-400e-be58-f84e0512985e)
97
+ - [Google/Fleurs Test Set](https://huggingface.co/datasets/google/fleurs)
98
+ - [IISc-MILE Test Set](https://www.openslr.org/127/)
99
+ - Babel Test Set
100
+
101
+ ## Training hyperparameters
102
 
103
  The following hyperparameters were used during training:
104
  - learning_rate: 0.75e-05
 
112
  - mixed_precision_training: True
113
 
114
  ## Acknowledgement
115
+ This work was done at [Speech Lab, IIT Madras](https://asr.iitm.ac.in/).
116
+
117
+ The compute resources for this work were funded by "Bhashini: National Language translation Mission" project of the Ministry of Electronics and Information Technology (MeitY), Government of India.