File size: 2,555 Bytes
a8a91ce 590a4dc a8a91ce 65aa48f a8a91ce 590a4dc a8a91ce 590a4dc a8a91ce c798b57 a8a91ce b2e33cd a8a91ce badd187 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
---
license: apache-2.0
language:
- ru
library_name: transformers
pipeline_tag: automatic-speech-recognition
tags:
- asr
- Pytorch
- pruned
- audio
- automatic-speech-recognition
---
# Whisper-tiny-ru-pruned
## Model info
This is a pruned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) model with only russian tokens left.
Pruning was made without any fine-tuning. Method from [this post](https://medium.com/m/global-identity-2?redirectUrl=https%3A%2F%2Ftowardsdatascience.com%2Fhow-to-adapt-a-multilingual-t5-model-for-a-single-language-b9f94f3d9c90) was used.
## Size
Only 10% tokens was left including special whisper tokens (no language tokens except \<|ru|\> and \<|en|\>, no timestamp tokens), 200 most popular tokens from tokenizer and 4000 most popular Russian tokens computed by tokenization of russian text corpus.
Model size is 50% less then original whisper-tiny:
| | openai/whisper-tiny | waveletdeboshir/whisper-tiny-ru-pruned |
| :------ | :------ | :------ |
| n of parameters | 38 M | 19.4 M |
| n of parameters (with proj_out layer) | 57.6 M | 21 M |
| model file size | 151 Mb | 86 Mb |
| vocab_size | 51865 | 4207 |
## Usage
Model can be used as an original whisper:
```python
>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
>>> import torchaudio
>>> # load audio
>>> wav, sr = torchaudio.load("audio.wav")
>>> # load model and processor
>>> processor = WhisperProcessor.from_pretrained("waveletdeboshir/whisper-tiny-ru-pruned")
>>> model = WhisperForConditionalGeneration.from_pretrained("waveletdeboshir/whisper-tiny-ru-pruned")
>>> input_features = processor(wav[0], sampling_rate=sr, return_tensors="pt").input_features
>>> # generate token ids
>>> predicted_ids = model.generate(input_features)
>>> # decode token ids to text
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
['<|startoftranscript|><|ru|><|transcribe|><|notimestamps|> Начинаем работу.<|endoftext|>']
```
The context tokens can be removed from the start of the transcription by setting `skip_special_tokens=True`.
## Other pruned whisper models
* [waveletdeboshir/whisper-base-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-base-ru-pruned)
* [waveletdeboshir/whisper-small-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-small-ru-pruned)
## Metrics
TODO
You can fine-tune this model on your data to achive better performance.
## Colab for vocab pruning
Check https://github.com/waveletdeboshir/whisper-lang-remover |