---
base_model:
- openai/whisper-small
language:
- en
license: mit
metrics:
- wer
---

This is model is a finetune of the [openai/whisper-small](https://huggingface.co/openai/whisper-small) model using approximately 750 hours of general conversational audio from Part 3 of the [National Speech Corpus](https://www.imda.gov.sg/how-we-can-help/national-speech-corpus). These are the final results on the evaluation set (~95 hours of audio):
- Validation Loss: 0.386770
- WER: 14.257934