---
library_name: transformers
license: apache-2.0
base_model: openai/whisper-large-v3-turbo
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: whisper-large-v3-turbo-ft-btb-cv-cy
  results: []
datasets:
- techiaith/banc-trawsgrifiadau-bangor
- techiaith/commonvoice_18_0_cy
language:
- cy
pipeline_tag: automatic-speech-recognition
---


# whisper-large-v3-turbo-ft-btb-cv-cy

This model is a version of [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) finedtuned with 
transcriptions of Welsh language spontaneous speech [Banc Trawsgrifiadau Bangor (btb)](https://huggingface.co/datasets/techiaith/banc-trawsgrifiadau-bangor)
ac well as recordings of read speach from [Welsh Common Voice version 18 (cv)](https://huggingface.co/datasets/techiaith/commonvoice_18_0_cy) 
for additional training. 

The Whisper large-v3-turbo pre-trained model is a finetuned version of a pruned Whisper large-v3. In other words, this model is the 
same model as [techiaith/whisper-large-v3-ft-btb-cv-cy](https://huggingface.co/techiaith/whisper-large-v3-ft-btb-cv-cy), 
except that the number of decoding layers have been reduced. As a result, the model is way faster, at the expense
of a minor quality degradation. 

It achieves the following results on the [Banc Trawsgrifiadau Bangor'r test set](https://huggingface.co/datasets/techiaith/banc-trawsgrifiadau-bangor/viewer/default/test)

- WER: 30.27
- CER: 11.14

As such this model is suitable for faster verbatim transcribing of spontaneous or unplanned speech. 


## Usage

```python
from transformers import pipeline

transcriber = pipeline("automatic-speech-recognition", model="techiaith/whisper-large-v3-turbo-ft-btb-cv-cy")
result = transcriber(<path or url to soundfile>)
print (result)
```

`{'text': 'ymm, yn y pum mlynadd dwitha 'ma ti 'di... Ie. ...bod drw dipyn felly do?'}`