stats_data.py

import pandas as pd

# Metadata lookup for UI and filtering
SPEAKER_META = {
    "F01": {"Gender": "Female", "Severity": "Severe", "Dataset": "Torgo"},
    "F03": {"Gender": "Female", "Severity": "Mild", "Dataset": "Torgo"},
    "F04": {"Gender": "Female", "Severity": "Mild", "Dataset": "Torgo"},
    "M01": {"Gender": "Male", "Severity": "Moderate", "Dataset": "Torgo"},
    "M02": {"Gender": "Male", "Severity": "Mild", "Dataset": "Torgo"},
    "M03": {"Gender": "Male", "Severity": "Mild", "Dataset": "Torgo"},
    "M04": {"Gender": "Male", "Severity": "Moderate", "Dataset": "Torgo"},
    "M05": {"Gender": "Male", "Severity": "Severe", "Dataset": "Torgo"},
    "F02 (UA)": {"Gender": "Female", "Severity": "Severe", "Dataset": "UA-Speech"}
}

def get_indomain_breakdown():
    # Performance on Seen Torgo Speakers
    data = {
        "Speaker": ["M05", "F01", "M01", "M04", "M02", "M03", "F03", "F04"],
        "Severity": ["Severe", "Severe", "Moderate", "Moderate", "Mild", "Mild", "Mild", "Mild"],
        "Whisper Tiny": [12.1, 12.6, 32.7, 31.8, 62.1, 58.4, 61.2, 59.1],
        "5K Pure Model": [33.1, 34.2, 47.2, 45.6, 84.5, 81.8, 83.5, 82.8],
        "10K Triple-Mix": ["25.4%", "24.1% (LOSO)", "44.1%", "41.2%", "79.1%", "77.5%", "79.0%", "78.2%"]
    }
    df = pd.DataFrame(data)
    
    # Calculate Relative Gain from the best model (usually 5K for in-domain)
    df["Relative Gain (Best)"] = (((df["5K Pure Model"] - df["Whisper Tiny"]) / df["Whisper Tiny"]) * 100).round(1)
    
    # Formatting
    df["Whisper Tiny"] = df["Whisper Tiny"].astype(str) + "%"
    df["5K Pure Model"] = df["5K Pure Model"].astype(str) + "%"
    df["Relative Gain (Best)"] = "+" + df["Relative Gain (Best)"].astype(str) + "%"
    return df

def get_experimental_summary():
    # Summary of the three primary research conditions
    data = {
        "Condition": ["In-Domain (Seen Torgo)", "LOSO (Unseen Torgo F01)", "Zero-Shot (UA-Speech F02)"],
        "Whisper Tiny": [41.50, 12.38, 4.33],
        "5K Pure Model": [58.77, "N/A", 6.19],
        "10K Triple-Mix": [54.67, 24.76, 5.98]
    }
    df = pd.DataFrame(data)
    
    # Calculate Best Relative Gain
    best_scores = [58.77, 24.76, 6.19]
    whisper_base = [41.50, 12.38, 4.33]
    gains = [f"+{round(((b-w)/w)*100, 1)}%" for b, w in zip(best_scores, whisper_base)]
    df["Relative Gain (Best)"] = gains

    # Formatting for display
    df["Whisper Tiny"] = df["Whisper Tiny"].astype(str) + "%"
    df["5K Pure Model"] = df["5K Pure Model"].apply(lambda x: f"{x}%" if x != "N/A" else x)
    df["10K Triple-Mix"] = df["10K Triple-Mix"].astype(str) + "%"
    
    return df