Downloads · 30 days
10
20% of all-time downloads
Whisper-Small-Khmer-v3
phonsobon/Whisper-Small-Khmer-v3
Whisper-Small-Khmer-v3 is a automatic speech recognition model from phonsobon. Use it when you need speech turned into text. The card lists the license as apache-2.0.
This is a continued fine-tune of Vira21/Whisper-Small-Khmer (itself a Khmer fine-tune of openai/whisper-small), trained further on additional Khmer speech data.
All-time downloads
50
Public
Parameters
242M
2.9 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt1.9 GB · 67%
At a glance
- Task
- Automatic Speech Recognition
- License
- apache-2.0
- Model type
- whisper
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026
- SHA
- ecdf9dc9
Base models
Model card
From the Hugging Face model README
Whisper-Small-Khmer-v3
This is a continued fine-tune of Vira21/Whisper-Small-Khmer
(itself a Khmer fine-tune of openai/whisper-small), trained further on additional Khmer speech data.
This checkpoint corresponds to checkpoint-1600 from further training in
phonsobon/Whisper-Small-Khmer-v2.
Training data
Training setup
- Base checkpoint:
Vira21/Whisper-Small-Khmer(WER 0.3763 on its original eval set) - Hardware: Kaggle free-tier GPU (T4 x2)
- Checkpoint: step 1600
Usage
from transformers import pipeline
pipe = pipeline("automatic-speech-recognition", model="phonsobon/Whisper-Small-Khmer-v3")
result = pipe("path/to/audio.wav")
print(result["text"])
Intended use
Khmer-language automatic speech recognition. Built as part of Khmer NLP tooling work for the Chaktomuk Managements (CTM) system.
Limitations
- Evaluated only on the training data mix above; WER on held-out/unseen domains may differ.
- Inherits any biases/limitations of the base Whisper-small architecture and the Vira21 checkpoint.
- Task
- Automatic Speech Recognition
- Type
- whisper
- License
- apache-2.0
- Languages
- km
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026