app.py

import pesq
import numpy as np
import soundfile as sf
import gradio as gr

def calculate_pesq(reference_audio, degraded_audio):
    # Extract audio data and sample rates from Gradio's Audio widget output
    ref_signal, ref_sample_rate = reference_audio[0], reference_audio[1]
    deg_signal, deg_sample_rate = degraded_audio[0], degraded_audio[1]

    # Convert audio data to the same sample rate if needed
    if ref_sample_rate != deg_sample_rate:
        from scipy.signal import resample
        min_length = min(len(ref_signal), len(deg_signal))
        ref_signal = resample(ref_signal[:min_length], len(ref_signal))
        deg_signal = resample(deg_signal[:min_length], len(deg_signal))

    pesq_score = pesq.pesq(ref_signal, deg_signal, ref_sample_rate)
    return pesq_score

def calculate_snr(reference_audio, degraded_audio):
    ref_signal, _ = reference_audio
    deg_signal, _ = degraded_audio

    noise = ref_signal - deg_signal
    snr = 10 * np.log10(np.sum(ref_signal**2) / np.sum(noise**2))
    return snr

def calculate_audio_difference(reference_audio, degraded_audio):
    ref_signal, ref_sample_rate = reference_audio[0], reference_audio[1]
    deg_signal, deg_sample_rate = degraded_audio[0], degraded_audio[1]

    if ref_sample_rate != deg_sample_rate:
        from scipy.signal import resample
        min_length = min(len(ref_signal), len(deg_signal))
        ref_signal = resample(ref_signal[:min_length], len(ref_signal))
        deg_signal = resample(deg_signal[:min_length], len(deg_signal))

    diff_count = np.count_nonzero(ref_signal != deg_signal)
    total_samples = len(ref_signal)
    difference_percentage = (diff_count / total_samples) * 100

    return difference_percentage

def audio_quality_interface(reference_audio, degraded_audio):
    pesq_score = calculate_pesq(reference_audio, degraded_audio)
    snr = calculate_snr(reference_audio, degraded_audio)
    difference_percentage = calculate_audio_difference(reference_audio, degraded_audio)

    result = f"Risultati:\n\nPunteggio PESQ: {pesq_score:.2f}\nSNR: {snr:.2f}\nPercentuale di differenza: {difference_percentage:.2f}%"
    return result

iface = gr.Interface(fn=audio_quality_interface,
                     inputs=[gr.Audio(label="Audio di riferimento"), gr.Audio(label="Audio degradato")],
                     outputs=gr.Textbox(),
                     layout="vertical",
                     live=True)

iface.launch()