herurg/whisper-tiny-minds14-en-us-audio-course

Whisper Tiny fine-tuned on MINDS-14 en-US

This model is a fine-tuned version of openai/whisper-tiny on the American English (en-US) subset of PolyAI/minds14 for the Hugging Face Audio Course Unit 5 assignment.

It achieves the following results on the evaluation set:

• Loss: 0.625578
• Wer Ortho: 0.326342
• Wer: 0.322904

Hugging Face Audio Course

Assignment requirements:

• Base model: openai/whisper-tiny
• Dataset: PolyAI/minds14, configuration en-US
• First 450 samples used for training
• Remaining samples used for evaluation
• Preprocessing performed with num_proc=1
• Task: automatic-speech-recognition
• Passing criterion: normalized WER < 0.37

Result:

Wer: 0.322904

✅ Passing threshold satisfied.