from transformers import Wav2Vec2ForCTC, AutoProcessor
import torch

processor = AutoProcessor.from_pretrained("Hassen80/hassaniya-mms-asr")
model = Wav2Vec2ForCTC.from_pretrained("Hassen80/hassaniya-mms-asr")

# Process audio (16kHz, mono)
inputs = processor(audio_array, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits
predicted_ids = torch.argmax(logits, dim=-1)
text = processor.batch_decode(predicted_ids)[0]