MWirelabs/ne-asr

A multilingual automatic speech recognition (ASR) model for eight Northeast Indian languages, fine-tuned from openai/whisper-medium.

Languages

Language	Code	Test WER
Khasi	kha	16.89%
Garo	grt	9.31%
Mizo	lus	23.85%
Nagamese	nag	49.13%
Kokborok	trp	44.79%
Assamese	asm	20.98%
Chakma	ccp	54.25%
Wancho	wao	68.37%
Overall		36.06%

Training Data

• Vaani (ARTPARK-IISc/Vaani-transcription-part): 121,960 training samples across 7 languages
• Proprietary MWire corpus: 28,524 training samples across 5 languages (Khasi, Garo, Mizo, Nagamese, Kokborok)
• Total: 150,483 training samples

Training Details

• Base model: openai/whisper-medium
• Learning rate: 1e-5 with 500 warmup steps
• Steps: 8,000
• Batch size: 16 (gradient accumulation 2, effective 32)
• Mixed precision: fp16
• Language token: welsh proxy for all languages except Assamese

Usage

from transformers import WhisperProcessor, WhisperForConditionalGeneration
import soundfile as sf
import torch

processor = WhisperProcessor.from_pretrained("MWirelabs/ne-asr")
model = WhisperForConditionalGeneration.from_pretrained("MWirelabs/ne-asr")

audio, sr = sf.read("audio.wav")
inputs = processor(audio, sampling_rate=16000, return_tensors="pt")

# Force language token (use "welsh" for all except Assamese)
forced_ids = processor.get_decoder_prompt_ids(language="welsh", task="transcribe")
with torch.no_grad():
    predicted_ids = model.generate(inputs.input_features, forced_decoder_ids=forced_ids)

transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
print(transcription)

Citation

Paper forthcoming. If you use this model, please cite:

@misc{mwirelabs2026nearsr, title={NE-MultiSpeech: Multilingual ASR for Northeast Indian Languages}, author={MWire Labs}, year={2026} }

License

CC-BY-4.0. Developed by MWire Labs, Shillong, Meghalaya.