app.py

import gradio as gr
import numpy as np
import librosa
import matplotlib.pyplot as plt
import io
from PIL import Image
from scipy.stats import kurtosis, skew

def analyze_audio(audio_file, snr, rms, peak, thd, spectral, mfcc_check):
    if audio_file is None:
        return None, "Please upload an audio file"

    try:
        # Load audio
        y, sr = librosa.load(audio_file, sr=None, mono=True)

        if len(y) < sr * 0.1:
            return None, "Audio file too short"

        # Calculate metrics
        results = []

        # Basic metrics
        noise_floor = np.percentile(np.abs(y), 5)
        signal_power = np.mean(y ** 2)
        noise_power = noise_floor ** 2
        snr_val = 10 * np.log10(signal_power / (noise_power + 1e-10))

        rms_val = np.sqrt(np.mean(y ** 2))
        peak_val = 20 * np.log10(np.max(np.abs(y)) + 1e-10)

        # THD
        fft = np.abs(np.fft.rfft(y))
        if len(fft) > 1:
            fundamental_idx = np.argmax(fft[1:]) + 1
            fundamental_power = fft[fundamental_idx] ** 2
            harmonic_indices = [i * fundamental_idx for i in range(2, 6) if i * fundamental_idx < len(fft)]
            harmonic_power = np.sum([fft[idx] ** 2 for idx in harmonic_indices])
            thd_val = np.sqrt(harmonic_power / (fundamental_power + 1e-10)) * 100
        else:
            thd_val = 0.0

        # Spectral
        spec_centroid = np.mean(librosa.feature.spectral_centroid(y=y, sr=sr)[0])

        # MFCC
        mfcc_feat = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
        mfcc_mean = np.mean(mfcc_feat)

        # Build output
        output = "## Audio Quality Metrics\n\n"

        if snr:
            output += f"**SNR:** {snr_val:.2f} dB\n"
        if rms:
            output += f"**RMS Energy:** {rms_val:.4f}\n"
        if peak:
            output += f"**Peak Level:** {peak_val:.2f} dB\n"
        if thd:
            output += f"**THD:** {thd_val:.2f}%\n"
        if spectral:
            output += f"**Spectral Centroid:** {spec_centroid:.2f} Hz\n"
        if mfcc_check:
            output += f"**MFCC Mean:** {mfcc_mean:.4f}\n"

        # Create visualization
        fig, axes = plt.subplots(2, 1, figsize=(10, 6))

        # Waveform
        time = np.linspace(0, len(y) / sr, len(y))
        axes[0].plot(time, y, linewidth=0.5)
        axes[0].set_title('Waveform')
        axes[0].set_xlabel('Time (s)')
        axes[0].grid(True, alpha=0.3)

        # Spectrogram
        D = librosa.amplitude_to_db(np.abs(librosa.stft(y)), ref=np.max)
        librosa.display.specshow(D, sr=sr, x_axis='time', y_axis='hz', ax=axes[1])
        axes[1].set_title('Spectrogram')

        plt.tight_layout()

        buf = io.BytesIO()
        plt.savefig(buf, format='png', dpi=100, bbox_inches='tight')
        buf.seek(0)
        img = Image.open(buf)
        plt.close()

        return img, output

    except Exception as e:
        return None, f"Error: {str(e)}"

# Create interface
iface = gr.Interface(
    fn=analyze_audio,
    inputs=[
        gr.Audio(label="Upload Audio File", type="filepath"),
        gr.Checkbox(label="SNR (Signal-to-Noise Ratio)", value=True),
        gr.Checkbox(label="RMS Energy", value=True),
        gr.Checkbox(label="Peak Level", value=True),
        gr.Checkbox(label="THD (Total Harmonic Distortion)", value=True),
        gr.Checkbox(label="Spectral Centroid", value=False),
        gr.Checkbox(label="MFCC", value=False),
    ],
    outputs=[
        gr.Image(label="Visualization", type="pil"),
        gr.Textbox(label="Metrics", lines=10)
    ],
    title="Audio Quality Assessment",
    description="Upload an audio file and select metrics to analyze"
)

iface.launch(share=True)