ritesh19180/maitri-emotion-models

MAITRI Emotion Models

Pre-trained models for the MAITRI (Multimodal AI Assistant for Psychological & Physical Well-Being Monitoring) system.

Models Included

File	Description	Framework
facial_expression_model_weights.h5	Face emotion classifier (7 emotions)	TensorFlow/Keras
speech_emotion_model.pkl	Speech emotion classifier (5 emotions)	scikit-learn

Inference Usage

Python — Download with huggingface_hub

from huggingface_hub import hf_hub_download

# Download face emotion model
face_model_path = hf_hub_download(
    repo_id="ritesh19180/maitri-emotion-models",
    filename="facial_expression_model_weights.h5"
)

# Download speech emotion model
speech_model_path = hf_hub_download(
    repo_id="ritesh19180/maitri-emotion-models",
    filename="speech_emotion_model.pkl"
)

Face Emotion Inference

from deepface import DeepFace

result = DeepFace.analyze(
    img_path="face.jpg",
    actions=["emotion"],
    enforce_detection=False
)
print(result[0]["dominant_emotion"])

Speech Emotion Inference

import pickle
import librosa
import numpy as np

with open(speech_model_path, "rb") as f:
    model = pickle.load(f)

audio, sr = librosa.load("audio.wav", sr=22050)
mfccs = np.mean(librosa.feature.mfcc(y=audio, sr=sr, n_mfcc=20), axis=1)
# ... extract remaining features ...
prediction = model.predict(features.reshape(1, -1))
print(prediction[0])

Emotion Classes

Face: happy, sad, angry, fear, disgust, surprise, neutral Speech: happy, sad, angry, fear, neutral

Citation

Part of the MAITRI project — a privacy-first, edge-computing AI system for mental health monitoring.