mms-1b-all-shn

facebook/mms-1b-all fine-tuned for Shan (shn) speech recognition (CTC, character vocabulary, target_lang="shn").

• Training mode: adapter, 10 epochs, best checkpoint by CER
• Data: NorHsangPha/Shan-ASR-Nova, shan-tts-dataset (Kaggle)
• License follows the base model (CC-BY-NC 4.0)

metric	value
eval CER (greedy)	4.50%
eval WER (greedy)	64.09%

WER is high because Shan word spacing is inconsistent - CER is the meaningful number.

Usage

import torch, librosa
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

proc = Wav2Vec2Processor.from_pretrained("k1ngtai/mms-1b-all-shn")
proc.tokenizer.set_target_lang("shn")
model = Wav2Vec2ForCTC.from_pretrained("k1ngtai/mms-1b-all-shn", target_lang="shn", ignore_mismatched_sizes=True).eval()

audio, _ = librosa.load("clip.wav", sr=16000)
inputs = proc(audio, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits
print(proc.decode(logits.argmax(-1)[0]))