wave2vec_demo.py

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

# Load Wav2Vec2 Model
MODEL_NAME = "facebook/wav2vec2-large-960h"
processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME)
model = Wav2Vec2ForCTC.from_pretrained(MODEL_NAME)

def transcribe(audio_file):
    """
    Transcribes speech from an uploaded audio file or live microphone input.
    """
    try:
        # Load and convert audio to 16kHz
        audio, rate = librosa.load(audio_file, sr=16000)

        # Convert audio to tensor format for Wav2Vec
        input_values = processor(audio, sampling_rate=16000, return_tensors="pt").input_values

        # Run the model for transcription
        with torch.no_grad():
            logits = model(input_values).logits

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

        return transcription

    except Exception as e:
        return "Error processing file"

# UI Build
interface = gr.Interface(
    fn=transcribe,  
    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", label="Speak or Upload Audio"),
    outputs="text",
    title="Wav2Vec2 Speech-to-Text Transcription",
    description="Speak into your microphone or upload an audio file to get an automatic transcription.",
    live=True  # Real-time microphone processing
)

interface.launch(share=True)