Trelis/whisper-small-atc-5s

Whisper Small Atc 5S

Fine-tuned Whisper model based on openai/whisper-small.

Training Results

Metric	Base Model	Fine-tuned
WER	55.54%	69.23%

Note: WER increased by 13.69% compared to base model

Training Details

• Base Model: openai/whisper-small
• Training Dataset: Trelis/atc-train-1gb-5s
• Train Loss: 0.3593
• Training Time: 3.6 minutes

Inference

from transformers import pipeline

asr = pipeline("automatic-speech-recognition", model="Trelis/whisper-small-atc-5s")
result = asr("path/to/audio.wav")
print(result["text"])

Training Logs

Full training logs are available in training_log.txt.

---

Fine-tuned using Trelis Studio