import os import re import sys import glob import numpy as np import pandas as pd import matplotlib matplotlib.use('Agg') import matplotlib.pyplot as plt # Ensure UTF-8 output sys.stdout.reconfigure(encoding='utf-8') BASE_DIR = r"c:\Users\ahmed\.gemini\antigravity\scratch\Allignments\PHONEME_TIMINGS_EACH_RECITER" # Safe filename conversion for Windows filesystem def sanitize_filename(name): # Replace characters not allowed in Windows filenames: \ / : * ? " < > | replacements = { ':': '_colon_', '/': '_slash_', '\\': '_bslash_', '*': '_star_', '?': '_qmark_', '"': '_quote_', '<': '_lt_', '>': '_gt_', '|': '_pipe_' } s = str(name) for char, rep in replacements.items(): s = s.replace(char, rep) return s def process_reciter(csv_path): filename = os.path.basename(csv_path) reciter_name = filename.replace('_phoneme_timestamps.csv', '') print(f"==================================================") print(f"Processing Reciter: {reciter_name}") print(f"==================================================") # Create Reciter Output Directory reciter_dir = os.path.join(BASE_DIR, reciter_name) step1_dir = os.path.join(reciter_dir, "step1_raw_histograms") step2_dir = os.path.join(reciter_dir, "step2_log_histograms") step3_dir = os.path.join(reciter_dir, "step3_boxplots_filtering") step4_dir = os.path.join(reciter_dir, "step4_filtered_histograms") for d in [step1_dir, step2_dir, step3_dir, step4_dir]: os.makedirs(d, exist_ok=True) df = pd.read_csv(csv_path) # Filter valid durations (> 0) df = df[df['duration'] > 0].copy() unique_phones = sorted(df['phone'].dropna().unique()) print(f"Total Unique Phonemes for {reciter_name}: {len(unique_phones)}") stats_list = [] # Configure matplotlib style plt.rcParams.update({ 'font.size': 10, 'axes.labelsize': 11, 'axes.titlesize': 12, 'figure.titlesize': 13, 'figure.autolayout': True }) for phone in unique_phones: phone_str = str(phone) safe_phone = sanitize_filename(phone_str) sub_df = df[df['phone'] == phone] t_raw = sub_df['duration'].values n_raw = len(t_raw) if n_raw == 0: continue # Step 2: Log Transformation y = ln(t) y_log = np.log(t_raw) # Step 3: IQR Outlier Filtering in Log Space q25 = np.percentile(y_log, 25) q75 = np.percentile(y_log, 75) iqr = q75 - q25 lower_bound_log = q25 - 1.5 * iqr upper_bound_log = q75 + 1.5 * iqr # In linear space cutoff lower_bound_linear = np.exp(lower_bound_log) upper_bound_linear = np.exp(upper_bound_log) # Filter mask in log space valid_mask = (y_log >= lower_bound_log) & (y_log <= upper_bound_log) # Step 4: Inverse Transform back to Linear Time t_filtered = t_raw[valid_mask] n_filtered = len(t_filtered) n_outliers = n_raw - n_filtered outlier_ratio = (n_outliers / n_raw) * 100.0 if n_raw > 0 else 0.0 # Step 5: Compute stats after filtering if n_filtered > 0: mean_filt = np.mean(t_filtered) median_filt = np.median(t_filtered) min_filt = np.min(t_filtered) max_filt = np.max(t_filtered) std_filt = np.std(t_filtered) else: mean_filt = median_filt = min_filt = max_filt = std_filt = np.nan stats_list.append({ 'reciter': reciter_name, 'phone': phone_str, 'raw_count': n_raw, 'filtered_count': n_filtered, 'outliers_removed': n_outliers, 'outlier_ratio_pct': round(outlier_ratio, 2), 'min_sec': round(min_filt, 4) if not np.isnan(min_filt) else None, 'max_sec': round(max_filt, 4) if not np.isnan(max_filt) else None, 'mean_sec': round(mean_filt, 4) if not np.isnan(mean_filt) else None, 'median_sec': round(median_filt, 4) if not np.isnan(median_filt) else None, 'std_sec': round(std_filt, 4) if not np.isnan(std_filt) else None, 'min_ms': round(min_filt * 1000, 1) if not np.isnan(min_filt) else None, 'max_ms': round(max_filt * 1000, 1) if not np.isnan(max_filt) else None, 'mean_ms': round(mean_filt * 1000, 1) if not np.isnan(mean_filt) else None, 'median_ms': round(median_filt * 1000, 1) if not np.isnan(median_filt) else None, 'lower_bound_cutoff_ms': round(lower_bound_linear * 1000, 1), 'upper_bound_cutoff_ms': round(upper_bound_linear * 1000, 1) }) # ------------------------------------------------------------- # Plot Step 1: Raw Histogram (Linear Time) # ------------------------------------------------------------- fig, ax = plt.subplots(figsize=(6, 4), dpi=100) bins = min(50, max(10, int(np.sqrt(n_raw)))) ax.hist(t_raw * 1000, bins=bins, color='#2b5c8f', edgecolor='black', alpha=0.75) ax.set_title(f"Step 1: Raw Timing Histogram\nReciter: {reciter_name} | Phoneme: [{phone_str}] (N={n_raw})") ax.set_xlabel("Duration (ms)") ax.set_ylabel("Frequency") ax.axvline(np.mean(t_raw) * 1000, color='red', linestyle='--', linewidth=1.5, label=f"Mean: {np.mean(t_raw)*1000:.1f} ms") ax.axvline(np.median(t_raw) * 1000, color='green', linestyle=':', linewidth=1.5, label=f"Median: {np.median(t_raw)*1000:.1f} ms") ax.legend(loc='upper right') ax.grid(axis='y', alpha=0.3) fig.tight_layout() fig.savefig(os.path.join(step1_dir, f"{safe_phone}_raw_hist.png")) plt.close(fig) # ------------------------------------------------------------- # Plot Step 2: Log Transform Histogram [ln(t)] # ------------------------------------------------------------- fig, ax = plt.subplots(figsize=(6, 4), dpi=100) ax.hist(y_log, bins=bins, color='#d97724', edgecolor='black', alpha=0.75) ax.set_title(f"Step 2: Log-Transformed Histogram [ln(t)]\nReciter: {reciter_name} | Phoneme: [{phone_str}]") ax.set_xlabel("ln(Duration in seconds)") ax.set_ylabel("Frequency") ax.axvline(lower_bound_log, color='purple', linestyle='--', linewidth=1.5, label=f"IQR Lower: {lower_bound_log:.2f}") ax.axvline(upper_bound_log, color='purple', linestyle='--', linewidth=1.5, label=f"IQR Upper: {upper_bound_log:.2f}") ax.legend(loc='upper right') ax.grid(axis='y', alpha=0.3) fig.tight_layout() fig.savefig(os.path.join(step2_dir, f"{safe_phone}_log_hist.png")) plt.close(fig) # ------------------------------------------------------------- # Plot Step 3: Box Plots Before vs After Outlier Filtering # ------------------------------------------------------------- fig, (ax_box1, ax_box2) = plt.subplots(1, 2, figsize=(8, 4.5), dpi=100, sharey=False) # Left: Linear Space boxplot (Before vs After) data_to_plot = [t_raw * 1000, t_filtered * 1000] ax_box1.boxplot(data_to_plot, tick_labels=['Raw (Before)', 'Filtered (After)'], patch_artist=True, boxprops=dict(facecolor='#8ecae6', color='#023047'), medianprops=dict(color='#d90429', linewidth=2), flierprops=dict(marker='o', markersize=4, markerfacecolor='red', alpha=0.5)) ax_box1.set_ylabel("Duration (ms)") ax_box1.set_title("Linear Time (ms)") ax_box1.grid(axis='y', alpha=0.3) # Right: Log Space boxplot with IQR cutoffs ax_box2.boxplot([y_log], tick_labels=['Log Space ln(t)'], patch_artist=True, boxprops=dict(facecolor='#ffb703', color='#fb8500'), medianprops=dict(color='#023047', linewidth=2), flierprops=dict(marker='x', markersize=5, markeredgecolor='purple', alpha=0.7)) ax_box2.axhline(lower_bound_log, color='red', linestyle='--', label=f'Lower Bound ({lower_bound_log:.2f})') ax_box2.axhline(upper_bound_log, color='red', linestyle='--', label=f'Upper Bound ({upper_bound_log:.2f})') ax_box2.set_ylabel("ln(t)") ax_box2.set_title("Log Space & IQR Bounds") ax_box2.legend(loc='lower right', fontsize=8) ax_box2.grid(axis='y', alpha=0.3) fig.suptitle(f"Step 3: Outlier Filtering Comparison | [{phone_str}] ({reciter_name})\nFiltered: {n_outliers} outliers ({outlier_ratio:.1f}%)", fontsize=11) fig.tight_layout() fig.savefig(os.path.join(step3_dir, f"{safe_phone}_boxplot.png")) plt.close(fig) # ------------------------------------------------------------- # Plot Step 4: Filtered Histogram (Linear Time) # ------------------------------------------------------------- fig, ax = plt.subplots(figsize=(6, 4), dpi=100) bins_filt = min(40, max(10, int(np.sqrt(n_filtered)))) ax.hist(t_filtered * 1000, bins=bins_filt, color='#2a9d8f', edgecolor='black', alpha=0.8) ax.set_title(f"Step 4: Filtered Timing Histogram (Linear Time)\nReciter: {reciter_name} | Phoneme: [{phone_str}] (N={n_filtered})") ax.set_xlabel("Duration (ms)") ax.set_ylabel("Frequency") if n_filtered > 0: ax.axvline(mean_filt * 1000, color='#e76f51', linestyle='--', linewidth=1.5, label=f"Mean: {mean_filt*1000:.1f} ms") ax.axvline(median_filt * 1000, color='#264653', linestyle=':', linewidth=1.5, label=f"Median: {median_filt*1000:.1f} ms") ax.axvline(min_filt * 1000, color='#457b9d', linestyle='-', linewidth=1.0, label=f"Min: {min_filt*1000:.1f} ms") ax.axvline(max_filt * 1000, color='#457b9d', linestyle='-', linewidth=1.0, label=f"Max: {max_filt*1000:.1f} ms") ax.legend(loc='upper right') ax.grid(axis='y', alpha=0.3) fig.tight_layout() fig.savefig(os.path.join(step4_dir, f"{safe_phone}_filtered_hist.png")) plt.close(fig) # Save CSV Summary for this reciter stats_df = pd.DataFrame(stats_list) summary_csv_path = os.path.join(reciter_dir, f"{reciter_name}_phase1_summary.csv") stats_df.to_csv(summary_csv_path, index=False, encoding='utf-8-sig') print(f"Saved summary CSV for {reciter_name}: {summary_csv_path}") print(f"Total Phonemes processed: {len(stats_df)}") print(f"Overall average outlier ratio: {stats_df['outlier_ratio_pct'].mean():.2f}%\n") return stats_df def main(): csv_files = glob.glob(os.path.join(BASE_DIR, "*_phoneme_timestamps.csv")) print(f"Found {len(csv_files)} reciter CSVs:") for f in csv_files: print(" -", os.path.basename(f)) all_summaries = [] for csv_file in csv_files: reciter_df = process_reciter(csv_file) all_summaries.append(reciter_df) master_summary = pd.concat(all_summaries, ignore_index=True) master_csv_path = os.path.join(BASE_DIR, "all_reciters_phase1_summary.csv") master_summary.to_csv(master_csv_path, index=False, encoding='utf-8-sig') print(f"\n Master summary saved across all reciters to: {master_csv_path}") if __name__ == '__main__': main()