whisper-eval-serbian 0.0.31

An evaluation framework for Serbian Whisper models.

AiDA Whisper Evaluation Framework (Serbian)

An evaluation framework for Serbian Whisper models.

Whisper Evaluator 🎤

A simple, modular framework to evaluate fine-tuned Whisper models in Python notebooks.

This library allows you to easily run evaluations on any dataset from the Hugging Face Hub using a simple configuration dictionary. It calculates a comprehensive set of metrics, including WER, CER, BLEU, and ROUGE, and automatically logs all results to a file.

Installation

You can install the library directly from GitHub for latest updates and features. Make sure you have git installed on your system.

pip install git+[https://github.com/datatab/whisper-evaluator.git](https://github.com/datatab/whisper-evaluator.git)

Quickstart

Using the library in a Google Colab or Jupyter Notebook is straightforward.

from whisper_evaluator import Evaluator
import json

# 1. Define your evaluation configuration
config = {
    "model_args": {
        "name_or_path": "openai/whisper-large-v2", # Your fine-tuned model ID
        "device": "cuda"
    },
    "task_args": {
        "dataset_name": "mozilla-foundation/common_voice_11_0",
        "dataset_subset": "sr", # Serbian language
        "dataset_split": "test[:20]", # Use the first 20 samples for a quick demo
        "audio_column": "audio",
        "text_column": "sentence"
    }
}

# 2. Initialize the evaluator
evaluator = Evaluator(config=config)

# 3. Run the evaluation (logs to 'evaluation_log.txt' by default)
detailed_results, metrics = evaluator.run()

# 4. Analyze the results
print("\n--- Final Metrics ---")
# Pretty print the metrics dictionary
print(json.dumps(metrics, indent=2))

print("\n--- Sample of evaluation details ---")
# Print the first 3 results from the list
for i, result in enumerate(detailed_results[:3]):
    print(f"\n--- Example {i+1} ---")
    print(f"Reference:  {result['reference']}")
    print(f"Prediction: {result['prediction']}")

Project Setup

Follow these steps to set up the AiDA-Whisper-Eval project.

---

Using Conda

1. Create a new Conda environment

conda create --name aida python=3.12 -y
conda activate aida

2. Install Poetry

pip install poetry

3. Install project dependencies

Navigate to the project's root directory and run:

poetry install

---

Using Plain Python

1. Create and activate a virtual environment

python -m venv venv

# On Linux/macOS
source venv/bin/activate

# On Windows
.\venv\Scripts\activate

2. Upgrade pip and install Poetry

pip install --upgrade pip
pip install poetry

3. Install project dependencies

From the project's root directory, run:

poetry install

pip install pre-commit

4. Set up pre-commit hooks

poetry run pre-commit install

---

Verifying Installation

Check installation by running tests:

# On Linux/macOS
make test

# On Windows
poetry run pytest

Your setup is complete!