tts/scripts/analysis/analyze_accent_verification.py

#!/usr/bin/env python3
"""
Analyze accent verification files to check for distinct accent characteristics
"""

import os
import numpy as np
import soundfile as sf
import scipy.signal
from scipy.stats import skew, kurtosis

# Paths
WORKSPACE = "/root/tts"
ACCENT_DIR = os.path.join(WORKSPACE, "accent_verification")

def calculate_rms(audio_data):
    """Calculate RMS energy"""
    return np.sqrt(np.mean(audio_data**2))

def calculate_peak_amplitude(audio_data):
    """Calculate peak amplitude"""
    return np.max(np.abs(audio_data))

def calculate_zero_crossing_rate(audio_data):
    """Calculate zero crossing rate"""
    return np.mean(np.abs(np.diff(np.sign(audio_data))))

def calculate_spectral_centroid(audio_data, sample_rate):
    """Calculate spectral centroid"""
    frequencies, times, Sxx = scipy.signal.spectrogram(audio_data, sample_rate)
    if np.sum(Sxx) == 0:
        return 0
    spectral_centroid = np.sum(frequencies[:, np.newaxis] * Sxx) / np.sum(Sxx)
    return spectral_centroid

def calculate_skewness(audio_data):
    """Calculate skewness"""
    return skew(audio_data)

def calculate_kurtosis(audio_data):
    """Calculate kurtosis"""
    return kurtosis(audio_data)

def analyze_audio_quality(audio_data, sample_rate, filename):
    """Analyze audio quality"""
    rms = calculate_rms(audio_data)
    peak = calculate_peak_amplitude(audio_data)
    zcr = calculate_zero_crossing_rate(audio_data)
    spectral_centroid = calculate_spectral_centroid(audio_data, sample_rate)
    skewness = calculate_skewness(audio_data)
    kurt = calculate_kurtosis(audio_data)
    
    # Quality scoring
    score = 0
    if 0.05 <= rms <= 0.3:
        score += 20
    if peak <= 1.0:
        score += 20
    if 0.05 <= zcr <= 0.3:
        score += 20
    if 400 <= spectral_centroid <= 3000:
        score += 20
    if -1 <= skewness <= 1:
        score += 10
    if kurt <= 10:
        score += 10
    
    return {
        'rms': rms,
        'peak': peak,
        'zcr': zcr,
        'spectral_centroid': spectral_centroid,
        'skewness': skewness,
        'kurtosis': kurt,
        'score': min(score, 100)
    }

def analyze_accent_verification():
    """Analyze accent verification files"""
    print("=" * 70)
    print("ANALYZING ACCENT VERIFICATION FILES")
    print("=" * 70)
    
    accent_files = []
    emotion_files = []
    
    # Get all files
    for filename in os.listdir(ACCENT_DIR):
        if filename.endswith('.wav'):
            file_path = os.path.join(ACCENT_DIR, filename)
            if 'accent' in filename:
                accent_files.append((filename, file_path))
            elif 'emotion' in filename:
                emotion_files.append((filename, file_path))
    
    # Analyze accent files
    print("\n🔊 ACCENT FILES ANALYSIS:")
    print("-" * 70)
    
    accent_stats = []
    for filename, file_path in accent_files:
        try:
            audio_data, sample_rate = sf.read(file_path)
            duration = len(audio_data) / sample_rate
            
            stats = analyze_audio_quality(audio_data, sample_rate, filename)
            
            accent_stats.append({
                'filename': filename,
                'duration': duration,
                'rms': stats['rms'],
                'zcr': stats['zcr'],
                'spectral_centroid': stats['spectral_centroid'],
                'score': stats['score']
            })
            
            print(f"✓ {filename}")
            print(f"  Duration: {duration:.2f}s, RMS: {stats['rms']:.4f}, ZCR: {stats['zcr']:.4f}, Centroid: {stats['spectral_centroid']:.1f}Hz, Score: {stats['score']}/100")
            print()
            
        except Exception as e:
            print(f"✗ {filename}: Error - {e}")
            print()
    
    # Analyze emotion files
    print("\n😊 EMOTION FILES ANALYSIS:")
    print("-" * 70)
    
    emotion_stats = []
    for filename, file_path in emotion_files:
        try:
            audio_data, sample_rate = sf.read(file_path)
            duration = len(audio_data) / sample_rate
            
            stats = analyze_audio_quality(audio_data, sample_rate, filename)
            
            emotion_stats.append({
                'filename': filename,
                'duration': duration,
                'rms': stats['rms'],
                'zcr': stats['zcr'],
                'spectral_centroid': stats['spectral_centroid'],
                'score': stats['score']
            })
            
            print(f"✓ {filename}")
            print(f"  Duration: {duration:.2f}s, RMS: {stats['rms']:.4f}, ZCR: {stats['zcr']:.4f}, Centroid: {stats['spectral_centroid']:.1f}Hz, Score: {stats['score']}/100")
            print()
            
        except Exception as e:
            print(f"✗ {filename}: Error - {e}")
            print()
    
    # Compare accent characteristics
    print("\n📊 ACCENT COMPARISON:")
    print("-" * 70)
    print("Filename                | Duration | RMS    | ZCR    | Centroid | Score")
    print("-" * 70)
    
    for stats in sorted(accent_stats, key=lambda x: x['filename']):
        print(f"{stats['filename']:24} | {stats['duration']:8.2f} | {stats['rms']:6.4f} | {stats['zcr']:6.4f} | {stats['spectral_centroid']:8.1f} | {stats['score']:5}")
    
    # Compare emotion characteristics
    print("\n📊 EMOTION COMPARISON:")
    print("-" * 70)
    print("Filename                | Duration | RMS    | ZCR    | Centroid | Score")
    print("-" * 70)
    
    for stats in sorted(emotion_stats, key=lambda x: x['filename']):
        print(f"{stats['filename']:24} | {stats['duration']:8.2f} | {stats['rms']:6.4f} | {stats['zcr']:6.4f} | {stats['spectral_centroid']:8.1f} | {stats['score']:5}")
    
    # Summary
    print("\n" + "=" * 70)
    print("SUMMARY")
    print("=" * 70)
    print(f"Total accent files: {len(accent_files)}")
    print(f"Total emotion files: {len(emotion_files)}")
    
    # Check if accents are distinct
    if len(accent_stats) >= 2:
        centroid_values = [s['spectral_centroid'] for s in accent_stats]
        centroid_std = np.std(centroid_values)
        zcr_values = [s['zcr'] for s in accent_stats]
        zcr_std = np.std(zcr_values)
        
        print(f"\nAccent distinctiveness metrics:")
        print(f"Spectral centroid std: {centroid_std:.2f}Hz (higher = more distinct)")
        print(f"Zero crossing rate std: {zcr_std:.4f} (higher = more distinct)")
        
        if centroid_std > 50 or zcr_std > 0.02:
            print("✅ Accents appear to be distinct based on acoustic features")
        else:
            print("⚠️  Accents may sound similar based on acoustic features")
    
    print("\n" + "=" * 70)

if __name__ == "__main__":
    analyze_accent_verification()