app.py

import gradio as gr
import torch
import librosa
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC

# Load pre-trained model and processor
MODEL_ID = "itsbib/wav2vec2-large-xls-r-300m-nepali-english-asr"
processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)

# Function to transcribe audio
def transcribe(audio):
    if isinstance(audio, tuple):  # If recorded, Gradio returns a tuple (file path, sample rate)
        audio = audio[0]  # Extract file path

    # Load and process audio
    audio_input, _ = librosa.load(audio, sr=16000)
    inputs = processor(audio_input, sampling_rate=16000, return_tensors="pt", padding=True)

    # Generate predictions
    with torch.no_grad():
        logits = model(inputs.input_values).logits

    # Decode predicted text
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)[0]

    return transcription

# Create Gradio interface
iface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath"),  # Allows both recording & file upload
    outputs="text",
    title="Nepali to English ASR",
    description="🎤 Record live or upload a Nepali audio file, and this model will transcribe it into English text."
)

iface.launch()