metadata

license: mit
datasets:
  - openslr/openslr
  - seanghay/km-speech-corpus
  - ylacombe/english_dialects
  - google/fleurs
language:
  - km
  - en
metrics:
  - wer
base_model:
  - openai/whisper-base
new_version: Vira21/Whisper-Base-KhmerV2
pipeline_tag: automatic-speech-recognition
library_name: transformers

Whisper-Base-KhmerV2

This model is a fine-tuned variant of openai/whisper-base, specifically adapted to enhance performance on diverse datasets. Designed to deliver improved transcription accuracy across multiple languages, including Khmer, it is fine-tuned with a focus on understanding the nuances of non-English languages and dialects.

Explore its capabilities in real-time transcription and multilingual support in the demo space: Whisper-Base-KhmerV2 Demo.

Metrics:
- WER (Word Error Rate): 0.4529
- Training Loss: 0.1012