vasista22 commited on
Commit
5661176
1 Parent(s): 5f954b9

inference_code_snippet_added

Browse files
Files changed (1) hide show
  1. README.md +56 -6
README.md CHANGED
@@ -31,14 +31,62 @@ should probably proofread and complete it, then remove this comment. -->
31
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Telugu data available from multiple publicly available ASR corpuses.
32
  It has been fine-tuned as a part of the Whisper fine-tuning sprint.
33
 
34
- ## Training and evaluation data at Speech Lab, IITM
35
 
36
- Training Data: CSTD IIIT-H ASR Corpus, ULCA ASR Corpus, Shrutilipi ASR Corpus, Microsoft Research Telugu Corpus (Train+Dev), Babel ASR Corpus, Google/Fleurs (Train+Dev) set.
37
- Evaluation Data: Babel Test, Microsoft Research Telugu Corpus Test, Google/Fleurs Test set, OpenSLR.
38
 
39
- ## Training procedure
40
 
41
- ### Training hyperparameters
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
42
 
43
  The following hyperparameters were used during training:
44
  - learning_rate: 5e-05
@@ -52,4 +100,6 @@ The following hyperparameters were used during training:
52
  - mixed_precision_training: True
53
 
54
  ## Acknowledgement
55
- 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.
 
 
 
31
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Telugu data available from multiple publicly available ASR corpuses.
32
  It has been fine-tuned as a part of the Whisper fine-tuning sprint.
33
 
34
+ **NOTE:** The code used to train this model is available for re-use in the [whisper-finetune](https://github.com/vasistalodagala/whisper-finetune) repository.
35
 
36
+ ## Usage
 
37
 
38
+ 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.
39
 
40
+ The same repository also provides the scripts for faster inference using whisper-jax.
41
+
42
+ In order to infer a single audio file using this model, the following code snippet can be used:
43
+
44
+ ```python
45
+ >>> import torch
46
+ >>> from transformers import pipeline
47
+
48
+ >>> # path to the audio file to be transcribed
49
+ >>> audio = "/path/to/audio.format"
50
+ >>> device = "cuda:0" if torch.cuda.is_available() else "cpu"
51
+
52
+ >>> transcribe = pipeline(task="automatic-speech-recognition", model="vasista22/whisper-telugu-tiny", chunk_length_s=30, device=device)
53
+ >>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="te", task="transcribe")
54
+
55
+ >>> print('Transcription: ', transcribe(audio)["text"])
56
+ ```
57
+
58
+ 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:
59
+
60
+ ```python
61
+ >>> import jax.numpy as jnp
62
+ >>> from whisper_jax import FlaxWhisperForConditionalGeneration, FlaxWhisperPipline
63
+
64
+ >>> # path to the audio file to be transcribed
65
+ >>> audio = "/path/to/audio.format"
66
+
67
+ >>> transcribe = FlaxWhisperPipline("vasista22/whisper-telugu-tiny", batch_size=16)
68
+ >>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="te", task="transcribe")
69
+
70
+ >>> print('Transcription: ', transcribe(audio)["text"])
71
+ ```
72
+
73
+ ## Training and evaluation data
74
+
75
+ Training Data:
76
+ - [CSTD IIIT-H ASR Corpus](https://asr.iiit.ac.in/cstd_ulca/)
77
+ - [ULCA ASR Corpus](https://github.com/Open-Speech-EkStep/ULCA-asr-dataset-corpus#telugu-labelled-total-duration-is-102593-hours)
78
+ - [Shrutilipi ASR Corpus](https://ai4bharat.org/shrutilipi)
79
+ - [Microsoft Speech Corpus (Indian Languages)](https://msropendata.com/datasets/7230b4b1-912d-400e-be58-f84e0512985e)
80
+ - [Google/Fleurs Train+Dev set](https://huggingface.co/datasets/google/fleurs)
81
+ - Babel ASR Corpus
82
+
83
+ Evaluation Data:
84
+ - [Microsoft Speech Corpus (Indian Languages) Test Set](https://msropendata.com/datasets/7230b4b1-912d-400e-be58-f84e0512985e)
85
+ - [Google/Fleurs Test Set](https://huggingface.co/datasets/google/fleurs)
86
+ - [OpenSLR](https://www.openslr.org/66/)
87
+ - Babel Test Set
88
+
89
+ ## Training hyperparameters
90
 
91
  The following hyperparameters were used during training:
92
  - learning_rate: 5e-05
 
100
  - mixed_precision_training: True
101
 
102
  ## Acknowledgement
103
+ This work was done at [Speech Lab, IIT Madras](https://asr.iitm.ac.in/).
104
+
105
+ 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.