hsaim/whisper-tiny-minds14-en-us-output

language:

• en license: apache-2.0 base_model: openai/whisper-tiny tags:
• automatic-speech-recognition
• whisper
• audio
• generated_from_trainer datasets:
• PolyAI/minds14 pipeline_tag: automatic-speech-recognition

---

Fine-tuned Whisper Tiny for MInDS-14 (en-US)

This model fine-tunes openai/whisper-tiny for automatic speech recognition on the American English (en-US) subset of the PolyAI/minds14 dataset.

Model description

Whisper Tiny is a compact encoder-decoder Transformer model for speech recognition. This version was fine-tuned to transcribe short English spoken queries from the MInDS-14 dataset.

Training data

• Dataset: PolyAI/minds14
• Configuration: en-US
• Training examples: first 450 examples
• Evaluation examples: remaining 113 examples
• Audio sampling rate: 16 kHz

Training procedure

Setting	Value
Base model	openai/whisper-tiny
Task	Automatic speech recognition
Language	English
Epochs	10
Learning rate	1e-5
Train batch size	4
Gradient accumulation steps	4
Effective batch size	16
Optimizer selection metric	Normalized WER

Evaluation results

The model was evaluated on the held-out 113 examples.

Metric	Score
Normalized WER	0.2993
Orthographic WER	0.2961

Lower Word Error Rate (WER) is better. A normalized WER of 0.2993 corresponds to approximately 29.93% word error.

Usage

from transformers import pipeline

pipe = pipeline(
    "automatic-speech-recognition",
    model="hsaim/whisper-tiny-minds14-en-us-output",
)

result = pipe("path/to/audio.wav")
print(result["text"])

Limitations

This is an educational fine-tuning run using a small dataset. It is intended for experimentation and may perform poorly on long recordings, noisy audio, different accents, languages other than English, or domains unlike MInDS-14 spoken queries.