thantzinphyo/Whisper-Tiny-Myanmar-Full-Fine-Tune

Whisper Tiny Myanmar (Direct End-to-End Fine-Tuned)

This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) trained directly on the [thantzinphyo/burmese-speech-refined-openslr-80](https://huggingface.co/datasets/thantzinphyo/burmese-speech-refined-openslr-80) dataset.

Benchmark Results (Best Checkpoint — Step 1200)

• —Best Validation WER: 51.7196%
• —Best Validation CER: 13.1144% (Character Accuracy: ~86.89%)
• —Training Loss: 0.000098 | Validation Loss: 0.2324 ---

How to Use (Python Inference)

import torch
from transformers import pipeline

pipe = pipeline(
    "automatic-speech-recognition",
    model="thantzinphyo/Whisper-Tiny-Myanmar-Full-Fine-Tune",
    device=0 if torch.cuda.is_available() else -1
)

result = pipe("your_burmese_audio.wav", generate_kwargs={"language": "my", "task": "transcribe"})
print("Transcribed Text:", result["text"]) # အခု ကတော့ သဘောတရားလေး ပဲ ပြောပြတာပါ

Full 1500 Steps Training History

Hyperparameters

• —Base Model: openai/whisper-tiny (100% End-to-End)
• —Effective Batch Size: 64 (16 per device × 4 gradient accumulation)
• —Learning Rate: 1.5e-4 with Cosine Annealing Scheduler
• —Warmup Steps: 150
• —Total Steps: 1500