| import os | |
| import numpy as np | |
| import cv2 | |
| import librosa | |
| import joblib | |
| from deepface import DeepFace | |
| import streamlit as st | |
| from collections import Counter | |
| from moviepy import VideoFileClip | |
| emotion_map = { | |
| 'angry': 0, | |
| 'disgust': 1, | |
| 'fear': 2, | |
| 'happy': 3, | |
| 'neutral': 4, | |
| 'sad': 5 | |
| } | |
| def split_video_into_frames_and_analyze_emotions(video_path, frame_rate=1): | |
| cap = cv2.VideoCapture(video_path) | |
| if not cap.isOpened(): | |
| st.error("Error: Could not open video.") | |
| return | |
| frame_count = 0 | |
| success, frame = cap.read() | |
| emotion_counter = Counter() | |
| while success: | |
| if frame_count % frame_rate == 0: | |
| try: | |
| analysis = DeepFace.analyze(frame, actions=['emotion']) | |
| if isinstance(analysis, list): | |
| for result in analysis: | |
| dominant_emotion = result['dominant_emotion'] | |
| emotion_counter[dominant_emotion] += 1 | |
| else: | |
| dominant_emotion = analysis['dominant_emotion'] | |
| emotion_counter[dominant_emotion] += 1 | |
| except Exception as e: | |
| pass | |
| success, frame = cap.read() | |
| frame_count += 1 | |
| cap.release() | |
| if emotion_counter: | |
| highest_occurring_emotion = emotion_counter.most_common(1)[0][0] | |
| else: | |
| highest_occurring_emotion = None | |
| return highest_occurring_emotion | |
| def extract_audio_from_video(video_path): | |
| video_clip = VideoFileClip(video_path) | |
| audio_path = "temp_audio.wav" | |
| video_clip.audio.write_audiofile(audio_path) | |
| audio_array, sr = librosa.load(audio_path, sr=None) | |
| os.remove(audio_path) | |
| return audio_array, sr | |
| def extract_features(audio_array, sr, max_length=100): | |
| try: | |
| mfccs = librosa.feature.mfcc(y=audio_array, sr=sr, n_mfcc=13) | |
| chroma = librosa.feature.chroma_stft(y=audio_array, sr=sr) | |
| spectral_contrast = librosa.feature.spectral_contrast(y=audio_array, sr=sr) | |
| features = np.vstack([mfccs, chroma, spectral_contrast]) | |
| if features.shape[1] < max_length: | |
| features = np.pad(features, ((0, 0), (0, max_length - features.shape[1])), mode='constant') | |
| elif features.shape[1] > max_length: | |
| features = features[:, :max_length] | |
| return features.T | |
| except Exception as e: | |
| st.error(f"Error extracting features from audio: {str(e)}") | |
| return None | |
| def main(): | |
| with open("style.css") as f: | |
| st.markdown(f"<style>{f.read()}</style>", unsafe_allow_html=True) | |
| st.title("Emotion Detection from Video") | |
| uploaded_file = st.file_uploader("Upload a video", type=["mp4"]) | |
| if uploaded_file is not None: | |
| video_path = "uploaded_video.mp4" | |
| with open(video_path, "wb") as f: | |
| f.write(uploaded_file.read()) | |
| st.write("Processing video...please wait") | |
| highest_emotion = split_video_into_frames_and_analyze_emotions(video_path) | |
| audio_array, sr = extract_audio_from_video(video_path) | |
| model_path = "SVMexec_modeltesting113.pkl" | |
| svm_model = joblib.load(model_path) | |
| scaler = joblib.load('scaler.pkl') | |
| features = extract_features(audio_array, sr) | |
| if features is not None: | |
| features_2d = features.reshape(1, -1) | |
| features_normalized = scaler.transform(features_2d) | |
| predicted_class = svm_model.predict(features_normalized)[0] | |
| emotion_labels = ['angry', 'disgust', 'fear', 'happy', 'neutral', 'sad'] | |
| predicted_emotion = emotion_labels[predicted_class] | |
| if highest_emotion == predicted_emotion: | |
| st.write(f"The person in the video is {predicted_emotion}.") | |
| else: | |
| st.write(f"The emotions from the frames and audio do not match, but the facial expression seems to be {highest_emotion}, while the audio emotion seems to be {predicted_emotion}.") | |
| else: | |
| st.write("Failed to extract features from the audio file.") | |
| if __name__ == "__main__": | |
| main() | |