shan-stt-v2

Whisper-small fine-tuned for Shan (shn) speech recognition, continued from innkl24/shan-stt-v1. Shan is decoded under Whisper's Thai language token (language="thai").

Training data: NorHsangPha/Shan-ASR-Nova

metric	value
eval WER (%)	48.90
eval CER (%)	3.28

from transformers import WhisperForConditionalGeneration, WhisperProcessor
import librosa, torch
proc = WhisperProcessor.from_pretrained("k1ngtai/shan-stt-v2")
model = WhisperForConditionalGeneration.from_pretrained("k1ngtai/shan-stt-v2")
audio, _ = librosa.load("clip.wav", sr=16000)
feats = proc.feature_extractor(audio, sampling_rate=16000, return_tensors="pt").input_features
ids = model.generate(feats, language="thai", task="transcribe")
print(proc.tokenizer.batch_decode(ids, skip_special_tokens=True)[0])