whisper-small-ru

Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

This is a fine-tuned version of the OpenAI Whisper-small model for ASR in Russian on the Golos Farfield dataset.

Model Details

• Base Model: openai/whisper-small
• Fine-tuning dataset: Golos Farfield
• Language: Russian
• Training Regime: The model was fine-tuned for 600 steps with a learning rate of 1e-5.

How to Get Started with the Model

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

You can use this model directly with the Hugging Face pipeline API:

from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", model="feelmadrain/whisper-small-ru")
result = pipe("your_audio.wav")
print(result["text"])

Results

<!-- This section describes the evaluation protocols and provides the results. -->

The model achieves a Word Error Rate (WER) of ~ 16.7 % on the validation data and ~ 20.1 % on the test data

{'test_loss': 0.2500078082084656, 'test_wer_ortho': 20.090670484371273, 'test_wer': 20.078740157480315, 'test_runtime': 606.858, 'test_samples_per_second': 3.156, 'test_steps_per_second': 0.789}