README.md

metadata

license: apache-2.0
language:
  - km
pipeline_tag: automatic-speech-recognition
library_name: transformers

Khmer-English Code-Switching ASR

This is a fine-tuned Wav2Vec2 XLS-R model for Khmer-English code-switching automatic speech recognition (ASR).

---

Model Details

• Base Model: facebook/wav2vec2-xls-r-300m
• Fine-tuning: Conducted on Khmer-English code-switched speech
• Training Duration: 8 hours
• Datasets Used:
  • seanghay/km-speech-corpus
  • google/fleurs
  • openslr/openslr
• Languages: Khmer (km), English (en)
• Evaluation Metrics:
  • Word Error Rate (WER): 51%
  • Character Error Rate (CER): 35%

---

Demo

Try it live on Hugging Face Spaces: Khmer-English CS-ASR Demo