app.py

import gradio as gr
import torch
import torchaudio
from transformers import AutoProcessor, Wav2Vec2ForCTC

MODEL_ID = "sb-x/mms-1b-bbl"

processor = AutoProcessor.from_pretrained(MODEL_ID)
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
model.eval()

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

def transcribe(audio):
    if audio is None:
        return ""

    sr, wav = audio

    wav = torch.tensor(wav).float()

    if wav.ndim > 1:
        wav = wav.mean(dim=1)

    if sr != 16000:
        wav = torchaudio.functional.resample(wav, sr, 16000)

    inputs = processor(
        wav.numpy(),
        sampling_rate=16000,
        return_tensors="pt"
    )
    inputs = {k: v.to(device) for k, v in inputs.items()}

    with torch.no_grad():
        logits = model(**inputs).logits

    pred_ids = torch.argmax(logits, dim=-1)
    return processor.batch_decode(pred_ids)[0]

demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="numpy"),
    outputs=gr.Textbox(label="Transcription",lines=10),
    title="MMS-1b-bbl ASR Demo",
    description="Fine-tuned MMS ASR model on bbl data from mozilla (CC BY-NC 4.0)"
)

demo.launch()