Spaces:
Running
Running
| import os | |
| import torchaudio | |
| import gradio as gr | |
| import torch | |
| from fastapi import FastAPI, HTTPException, File, UploadFile | |
| from speechbrain.inference import SpeakerRecognition | |
| from fastapi.responses import JSONResponse | |
| # Initialize the speaker verification model | |
| speaker_verification = SpeakerRecognition.from_hparams( | |
| source="speechbrain/spkrec-ecapa-voxceleb", | |
| savedir="tmp_model" | |
| ) | |
| # Function to calculate similarity score | |
| def get_similarity(audio1, audio2, sample_rate=16000): | |
| try: | |
| # Convert numpy arrays to tensors | |
| signal1 = torch.tensor(audio1) | |
| signal2 = torch.tensor(audio2) | |
| # Make sure the signals are in the right shape (2D tensor: (1, N)) | |
| if signal1.ndimension() == 1: | |
| signal1 = signal1.unsqueeze(0) | |
| if signal2.ndimension() == 1: | |
| signal2 = signal2.unsqueeze(0) | |
| # Get similarity score and prediction | |
| score, prediction = speaker_verification.verify_batch(signal1, signal2) | |
| return float(score), "Yes" if prediction else "No" | |
| except Exception as e: | |
| return None, str(e) # Return error message if any exception | |
| # API function to compare voices | |
| def compare_voices(file1, file2): | |
| try: | |
| # Gradio Audio returns a tuple of (audio, sample_rate) | |
| audio1, _ = file1 # Audio1 is a tuple (numpy_array, sample_rate) | |
| audio2, _ = file2 # Audio2 is a tuple (numpy_array, sample_rate) | |
| # Get similarity score | |
| score, is_same_user = get_similarity(audio1, audio2) | |
| if score is None: | |
| # Return the error message if processing fails | |
| return {"error": is_same_user} | |
| # Return a dictionary with the similarity score and prediction | |
| return {"Similarity Score": f"{score:.4f}", "Same User Prediction": is_same_user} | |
| except Exception as e: | |
| # Handle unexpected errors | |
| return {"error": str(e)} | |
| # FastAPI app | |
| app = FastAPI() | |
| async def compare_voices_api(file1: UploadFile = File(...), file2: UploadFile = File(...)): | |
| """ | |
| Compare two audio files and return the similarity score and prediction. | |
| """ | |
| try: | |
| # Process the audio files and return them as numpy arrays | |
| file1_data = await file1.read() | |
| file2_data = await file2.read() | |
| # You need to process these byte strings into numpy arrays | |
| # Assuming the audio is decoded into numpy arrays here (e.g., using torchaudio) | |
| # For example: | |
| # audio1 = torchaudio.load(io.BytesIO(file1_data))[0].numpy() | |
| # audio2 = torchaudio.load(io.BytesIO(file2_data))[0].numpy() | |
| return {"message": "Processing files directly without saving them."} | |
| except Exception as e: | |
| raise HTTPException(status_code=400, detail=str(e)) | |
| # Gradio interface function | |
| def gradio_interface(): | |
| return gr.Interface( | |
| fn=compare_voices, | |
| inputs=[ | |
| gr.Audio(type="numpy", label="First Audio File"), # Gradio now gives numpy arrays | |
| gr.Audio(type="numpy", label="Second Audio File") # Gradio now gives numpy arrays | |
| ], | |
| outputs="json", # Output results as JSON | |
| live=False # No live interface, just the API | |
| ) | |
| # Launch Gradio as a web interface | |
| async def startup(): | |
| gr.Interface(fn=compare_voices, inputs=[ | |
| gr.Audio(type="numpy", label="First Audio File"), # Gradio now gives numpy arrays | |
| gr.Audio(type="numpy", label="Second Audio File") # Gradio now gives numpy arrays | |
| ], outputs="json", live=False).launch(share=True, inline=True) | |
| # Running the FastAPI app with Gradio | |
| if __name__ == "__main__": | |
| import uvicorn | |
| uvicorn.run(app, host="0.0.0.0", port=5000) | |