|
--- |
|
pipeline_tag: automatic-speech-recognition |
|
datasets: |
|
- mozilla-foundation/common_voice_11_0 |
|
license: cc |
|
language: |
|
- rw |
|
metrics: |
|
- cer |
|
base_model: |
|
- openai/whisper-small |
|
tags: |
|
- STT |
|
- fine-tune-kinyarwanda |
|
- kinyarwanda |
|
--- |
|
# Model description |
|
|
|
This model is an openai's whisper-small model fine-tuned on the Kinyarwanda common-voice dataset. The Kinyarwanda language was added by fine-tuning on top of the Swahili language. |
|
It achieves a 24 WER. Currently, it does not provide Kinyarwanda-to-English translation. |
|
|
|
# Usage |
|
```python |
|
>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration |
|
>>> from datasets import load_dataset |
|
>>> import datasets |
|
>>> import torch |
|
>>> # load model and processor |
|
>>> processor = WhisperProcessor.from_pretrained("mbazaNLP/Whisper-Small-Kinyarwanda") |
|
>>> model = WhisperForConditionalGeneration.from_pretrained("mbazaNLP/Whisper-Small-Kinyarwanda") |
|
>>> ds = load_dataset("common_voice", "rw", split="test", streaming=True) |
|
>>> ds = ds.cast_column("audio", datasets.Audio(sampling_rate=16_000)) |
|
>>> input_speech = next(iter(ds))["audio"]["array"] |
|
>>> model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language = "sw", task = "transcribe") |
|
>>> input_features = processor(input_speech, return_tensors="pt").input_features |
|
>>> predicted_ids = model.generate(input_features) |
|
>>> transcription = processor.batch_decode(predicted_ids) |
|
['<|startoftranscript|><|sw|><|transcribe|><|notimestamps|>Abamugariye ku rugamba bafashwa kubona insimburangingo<|endoftext|>'] |
|
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens = True) |
|
['Abamugariye ku rugamba bafashwa kubona insimburangingo'] |
|
|
|
``` |
