app.py

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

# Load pre-trained model and processor
model_name = "facebook/wav2vec2-base-960h"
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)

def speech_to_text(audio):
    try:
        if audio is None:
            return "No audio file provided."

        # Load audio file
        waveform, rate = torchaudio.load(audio)
        
        # Ensure the audio is mono
        if waveform.shape[0] > 1:
            waveform = torch.mean(waveform, dim=0, keepdim=True)
        
        # Resample to 16000 Hz
        resampler = torchaudio.transforms.Resample(orig_freq=rate, new_freq=16000)
        waveform = resampler(waveform)
        
        # Tokenize the waveform
        inputs = processor(waveform.squeeze().numpy(), return_tensors="pt", sampling_rate=16000)
        
        # Perform inference
        with torch.no_grad():
            logits = model(**inputs).logits
        
        # Decode the output
        predicted_ids = torch.argmax(logits, dim=-1)
        transcription = processor.batch_decode(predicted_ids)[0]
        
        return transcription
    except Exception as e:
        return str(e)

# Create Gradio interface
iface = gr.Interface(
    fn=speech_to_text,
    inputs=gr.Audio(type="filepath", label="Input Audio"),
    outputs=gr.Textbox(label="Transcription"),
    live=True,
    title="Speech to Text",
    description="Speak into your microphone and get the transcribed text."
)

# Launch the interface
iface.launch()