---
license: apache-2.0
language:
- mk
base_model:
- openai/whisper-large-v3
library_name: speechbrain
---

# Fine-tuned whisper-large-v3 model for speech recognition in Macedonian

Authors:
1. Dejan Porjazovski
2. Ilina Jakimovska
3. Ordan Chukaliev
4. Nikola Stikov

This collaboration is part of the activities of the Center for Advanced Interdisciplinary Research (CAIR) at UKIM.

## Data used for training

The model is trained on around 115 hours of Macedonian speech.


## Model description

This model is a fine-tuned version of the large Whisper-v3 model. During fine-tuning, the encoder was kept frozen and only the decoder was optimized.


## Results

The results are reported on all the test sets combined.

WER: 11.76 \
CER: 5.18


## Usage

The model is developed using the [SpeechBrain](https://speechbrain.github.io) toolkit. To use it, you need to install SpeechBrain with:
```
pip install speechbrain
```
SpeechBrain relies on the Transformers library, therefore you need install it:
```
pip install transformers
```

An external `py_module_file=custom_interface.py` is used as an external Predictor class into this HF repos. We use the `foreign_class` function from `speechbrain.pretrained.interfaces` that allows you to load your custom model. 

```python
from speechbrain.inference.interfaces import foreign_class
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
asr_classifier = foreign_class(source="Macedonian-ASR/buki-whisper-2.0", pymodule_file="custom_interface.py", classname="ASR")
asr_classifier = asr_classifier.to(device)
predictions = asr_classifier.classify_file("audio_file.wav", device)
print(predictions)
```

## Training

To fine-tune this model, you need to run:
```
python train.py hyperparams.yaml
```

```train.py``` file contains the functions necessary for training the model and ```hyperparams.yaml``` contains the hyperparameters. For more details about training the model, refer to the [SpeechBrain](https://speechbrain.github.io) documentation.