AI for Mental Health

Depression Screening from Speech

Acoustic Features

import librosa
import numpy as np

class SpeechDepressionAnalyzer:
    def extract_features(self, audio_path, sr=16000):
        y, _ = librosa.load(audio_path, sr=sr)
        f0, _, _ = librosa.pyin(y, fmin=50, f0_max=500)
        mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
        intervals = librosa.effects.split(y, top_db=30)
        pause_durations = np.diff(intervals) / sr
        return {
            'f0_mean': np.nanmean(f0), 'f0_std': np.nanstd(f0),
            'mfcc_mean': np.mean(mfcc, axis=1),
            'avg_pause': np.mean(pause_durations)
        }

Anxiety Detection

Behavioral Markers

• Speech patterns: Increased rate, higher pitch variability
• Language use: More self-referential pronouns
• Physiological: Heart rate variability, skin conductance

Multi-Modal Assessment

Therapy Chatbots

CBT Framework

1. Identify negative thought patterns
2. Challenge cognitive distortions
3. Replace with balanced thoughts
4. Practice behavioral activation

PHQ-9 Scoring