FEAT: Refactor app for post-hoc file upload analysis

requirements.txt

@@ -1,9 +1,10 @@
 --extra-index-url https://download.pytorch.org/whl/cpu
 
-# Pin the main UI components to recent, stable versions
+# Core app and UI library
 streamlit==1.35.0
-streamlit-webrtc==0.47.0
-streamlit-autorefresh==1.0.1
+
+# Library for video/audio file handling
+moviepy
 
 # Pin ML/AI libraries to modern, known-good versions
 transformers==4.40.1

src/streamlit_app.py

@@ -1,41 +1,173 @@
+import os
 import streamlit as st
-from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration
-import av
+import numpy as np
+import torch
+import whisper
+from transformers import pipeline, AutoModelForAudioClassification, AutoFeatureExtractor
+from deepface import DeepFace
+import logging
+import soundfile as sf
+from scipy.io.wavfile import write as write_wav
+import tempfile
+from PIL import Image
+import cv2
+from moviepy.editor import VideoFileClip
+
+# Set home directories for model caching inside the app's writable directory
+os.environ['DEEPFACE_HOME'] = '/tmp/.deepface'
+os.environ['HF_HOME'] = '/tmp/huggingface'
 
 # --- Page Configuration ---
 st.set_page_config(
-    page_title="AffectLink Webcam Test",
-    page_icon="📸",
+    page_title="AffectLink Batch Demo",
+    page_icon="😊",
     layout="wide"
 )
 
-st.title("AffectLink: Webcam Feed Test")
-st.write("This is a minimal test to check if the webcam video stream can be established.")
-
-# --- WebRTC Configuration ---
-# Using a robust list of public STUN servers to help establish a connection
-RTC_CONFIGURATION = RTCConfiguration(
-    {"iceServers": [{"urls": [
-        "stun:stun.l.google.com:19302",
-        "stun:stun1.l.google.com:19302",
-        "stun:stun2.l.google.com:19302",
-        "stun:stun3.l.google.com:19302",
-        "stun:stun4.l.google.com:19302",
-    ]}]}
-)
+st.title("AffectLink: Post-Hoc Emotion Analysis")
+st.write("Upload a short video clip to analyze facial expressions, speech-to-text, and the emotional tone of the audio.")
 
-# A simple "pass-through" function that returns the frame without modification
-def video_frame_callback(frame: av.VideoFrame) -> av.VideoFrame:
-    return frame
-
-# --- WebRTC Streamer ---
-webrtc_streamer(
-    key="webcam-test",
-    mode=WebRtcMode.SENDRECV,
-    rtc_configuration=RTC_CONFIGURATION,
-    media_stream_constraints={"video": True, "audio": False},
-    video_frame_callback=video_frame_callback,
-    async_processing=True,
-)
+# --- Logger Configuration ---
+logging.basicConfig(level=logging.INFO)
+logging.getLogger('deepface').setLevel(logging.ERROR)
+logging.getLogger('huggingface_hub').setLevel(logging.WARNING)
+logging.getLogger('moviepy').setLevel(logging.ERROR)
+
+
+# --- Emotion Mappings ---
+UNIFIED_EMOTIONS = ['neutral', 'happy', 'sad', 'angry']
+TEXT_TO_UNIFIED = {
+    'neutral': 'neutral', 'joy': 'happy', 'sadness': 'sad', 'anger': 'angry',
+    'fear': None, 'surprise': None, 'disgust': None
+}
+SER_TO_UNIFIED = {
+    'neu': 'neutral', 'hap': 'happy', 'sad': 'sad', 'ang': 'angry'
+}
+AUDIO_SAMPLE_RATE = 16000
+
+# --- Model Loading ---
+@st.cache_resource
+def load_models():
+    with st.spinner("Loading AI models, this may take a moment..."):
+        whisper_model = whisper.load_model("base", download_root="/tmp/whisper_cache")
+        text_classifier = pipeline("text-classification", model="j-hartmann/emotion-english-distilroberta-base", top_k=None)
+        ser_model_name = "superb/hubert-large-superb-er"
+        ser_feature_extractor = AutoFeatureExtractor.from_pretrained(ser_model_name)
+        ser_model = AutoModelForAudioClassification.from_pretrained(ser_model_name)
+        return whisper_model, text_classifier, ser_model, ser_feature_extractor
+
+whisper_model, text_classifier, ser_model, ser_feature_extractor = load_models()
+
+
+# --- UI and Processing Logic ---
+uploaded_file = st.file_uploader("Choose a video file...", type=["mp4", "mov", "avi"])
+
+if uploaded_file is not None:
+    # Save the uploaded file to a temporary location
+    with tempfile.NamedTemporaryFile(delete=False, suffix='.mp4') as tfile:
+        tfile.write(uploaded_file.read())
+        temp_video_path = tfile.name
+
+    st.video(temp_video_path)
+
+    if st.button("Analyze Video"):
+        facial_analysis_results = []
+        audio_analysis_results = {}
+
+        # --- Video Processing for Facial Emotion ---
+        with st.spinner("Analyzing video for facial expressions..."):
+            try:
+                cap = cv2.VideoCapture(temp_video_path)
+                fps = cap.get(cv2.CAP_PROP_FPS)
+                frame_count = 0
+                while cap.isOpened():
+                    ret, frame = cap.read()
+                    if not ret:
+                        break
+                    
+                    # Process one frame per second
+                    if frame_count % int(fps) == 0:
+                        timestamp = frame_count / fps
+                        analysis = DeepFace.analyze(frame, actions=['emotion'], enforce_detection=False, silent=True)
+                        if isinstance(analysis, list) and len(analysis) > 0:
+                            dominant_emotion = analysis[0]['dominant_emotion']
+                            facial_analysis_results.append((timestamp, dominant_emotion.capitalize()))
+                    
+                    frame_count += 1
+                cap.release()
+            except Exception as e:
+                st.error(f"An error occurred during facial analysis: {e}")
+
+
+        # --- Audio Extraction and Processing ---
+        with st.spinner("Extracting and analyzing audio..."):
+            try:
+                # Extract audio using moviepy
+                video_clip = VideoFileClip(temp_video_path)
+                with tempfile.NamedTemporaryFile(delete=False, suffix='.wav') as taudio:
+                    video_clip.audio.write_audiofile(taudio.name, fps=AUDIO_SAMPLE_RATE, logger=None)
+                    temp_audio_path = taudio.name
+
+                # 1. Speech-to-Text (Whisper)
+                result = whisper_model.transcribe(temp_audio_path, fp16=False)
+                transcribed_text = result['text']
+                audio_analysis_results['Transcription'] = transcribed_text
+
+                # 2. Text-based Emotion
+                if transcribed_text:
+                    text_emotions = text_classifier(transcribed_text)[0]
+                    unified_text_scores = {e: 0.0 for e in UNIFIED_EMOTIONS}
+                    for emo in text_emotions:
+                        unified_emo = TEXT_TO_UNIFIED.get(emo['label'])
+                        if unified_emo:
+                            unified_text_scores[unified_emo] += emo['score']
+                    dominant_text_emotion = max(unified_text_scores, key=unified_text_scores.get)
+                    audio_analysis_results['Text Emotion'] = dominant_text_emotion.capitalize()
+                
+                # 3. Speech Emotion Recognition (SER)
+                audio_array, _ = sf.read(temp_audio_path)
+                inputs = ser_feature_extractor(audio_array, sampling_rate=AUDIO_SAMPLE_RATE, return_tensors="pt", padding=True)
+                with torch.no_grad():
+                    logits = ser_model(**inputs).logits
+                scores = torch.nn.functional.softmax(logits, dim=1).squeeze()
+                unified_ser_scores = {e: 0.0 for e in UNIFIED_EMOTIONS}
+                for i, score in enumerate(scores):
+                    raw_emo = ser_model.config.id2label[i]
+                    unified_emo = SER_TO_UNIFIED.get(raw_emo)
+                    if unified_emo:
+                        unified_ser_scores[unified_emo] += score.item()
+                dominant_ser_emotion = max(unified_ser_scores, key=unified_ser_scores.get)
+                audio_analysis_results['Speech Emotion'] = dominant_ser_emotion.capitalize()
+                
+                # Clean up temp audio file
+                os.unlink(temp_audio_path)
+
+            except Exception as e:
+                st.error(f"An error occurred during audio analysis: {e}")
+            finally:
+                video_clip.close()
+
+
+        # --- Display Results ---
+        st.header("Analysis Results")
+        col1, col2 = st.columns(2)
+
+        with col1:
+            st.subheader("Audio Analysis")
+            if audio_analysis_results:
+                st.write(f"**Transcription:** \"{audio_analysis_results.get('Transcription', 'N/A')}\"")
+                st.metric("Emotion from Text", audio_analysis_results.get('Text Emotion', 'N/A'))
+                st.metric("Emotion from Speech", audio_analysis_results.get('Speech Emotion', 'N/A'))
+            else:
+                st.write("No audio results to display.")
+
+        with col2:
+            st.subheader("Facial Expression Timeline")
+            if facial_analysis_results:
+                for timestamp, emotion in facial_analysis_results:
+                    st.write(f"**Time {int(timestamp // 60):02d}:{int(timestamp % 60):02d}:** {emotion}")
+            else:
+                st.write("No faces detected or video processing failed.")
 
-st.sidebar.info("This is a test page. All AI analysis and audio processing has been disabled.")
+    # Clean up temp video file
+    os.unlink(temp_video_path)

src/streamlit_app.py.old

@@ -1,199 +0,0 @@
-import os
-import streamlit as st
-
-# Set home directories for model caching inside the app's writable directory
-os.environ['DEEPFACE_HOME'] = '/tmp/.deepface'
-os.environ['HF_HOME'] = '/tmp/huggingface'
-
-from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration
-import av
-import numpy as np
-import torch
-import whisper
-from transformers import pipeline, AutoModelForAudioClassification, AutoFeatureExtractor
-from deepface import DeepFace
-import logging
-import queue
-import soundfile as sf
-from scipy.io.wavfile import write as write_wav
-import tempfile
-
-# --- Page Configuration ---
-st.set_page_config(
-    page_title="AffectLink Online Demo",
-    page_icon="😊",
-    layout="wide"
-)
-
-st.title("AffectLink: Real-time Emotion Recognition")
-st.write("This demo analyzes your facial expressions in real-time and processes short audio clips for speech and text-based emotion.")
-
-# --- Logger Configuration ---
-logging.basicConfig(level=logging.INFO)
-logging.getLogger('deepface').setLevel(logging.ERROR)
-logging.getLogger('huggingface_hub').setLevel(logging.WARNING)
-
-# --- Emotion Mappings ---
-UNIFIED_EMOTIONS = ['neutral', 'happy', 'sad', 'angry']
-FACIAL_TO_UNIFIED = {
-    'neutral': 'neutral', 'happy': 'happy', 'sad': 'sad', 'angry': 'angry',
-    'fear': None, 'surprise': None, 'disgust': None
-}
-TEXT_TO_UNIFIED = {
-    'neutral': 'neutral', 'joy': 'happy', 'sadness': 'sad', 'anger': 'angry',
-    'fear': None, 'surprise': None, 'disgust': None
-}
-SER_TO_UNIFIED = {
-    'neu': 'neutral', 'hap': 'happy', 'sad': 'sad', 'ang': 'angry'
-}
-AUDIO_SAMPLE_RATE = 16000
-
-# --- Model Loading ---
-@st.cache_resource
-def load_models():
-    with st.spinner("Loading AI models... This may take a moment on first run."):
-        whisper_model = whisper.load_model("base", download_root="/tmp/whisper_cache")
-        text_classifier = pipeline("text-classification", model="j-hartmann/emotion-english-distilroberta-base", top_k=None)
-        ser_model_name = "superb/hubert-large-superb-er"
-        ser_feature_extractor = AutoFeatureExtractor.from_pretrained(ser_model_name)
-        ser_model = AutoModelForAudioClassification.from_pretrained(ser_model_name)
-        return whisper_model, text_classifier, ser_model, ser_feature_extractor
-
-whisper_model, text_classifier, ser_model, ser_feature_extractor = load_models()
-
-# --- WebRTC and Video Processing ---
-# === THIS IS THE UPDATED CONFIGURATION ===
-RTC_CONFIGURATION = RTCConfiguration(
-    {"iceServers": [{"urls": [
-        "stun:stun.l.google.com:19302",
-        "stun:stun1.l.google.com:19302",
-        "stun:stun2.l.google.com:19302",
-        "stun:stun3.l.google.com:19302",
-        "stun:stun4.l.google.com:19302",
-    ]}]}
-)
-# ==========================================
-
-webrtc_ctx = webrtc_streamer(
-    key="affectlink-video",
-    mode=WebRtcMode.SENDRECV,
-    rtc_configuration=RTC_CONFIGURATION, # Use the new config
-    media_stream_constraints={"video": True, "audio": False},
-    async_processing=True,
-)
-
-if 'facial_emotion' not in st.session_state:
-    st.session_state.facial_emotion = "Neutral"
-if 'last_emotion_time' not in st.session_state:
-    st.session_state.last_emotion_time = 0
-
-def video_frame_callback(frame: av.VideoFrame) -> av.VideoFrame:
-    img = frame.to_ndarray(format="bgr24")
-    current_time = st.session_state.get('last_emotion_time', 0)
-    if torch.cuda.is_available() or (hasattr(st.session_state, 'last_emotion_time') and (torch.tensor(current_time).item() + 5 < torch.tensor(frame.time).item())):
-        try:
-            analysis = DeepFace.analyze(img, actions=['emotion'], enforce_detection=False, silent=True)
-            if isinstance(analysis, list) and len(analysis) > 0:
-                dominant_emotion = analysis[0]['dominant_emotion']
-                st.session_state.facial_emotion = dominant_emotion.capitalize()
-            else:
-                st.session_state.facial_emotion = "Unknown"
-        except Exception as e:
-            logging.error(f"DeepFace analysis failed: {e}")
-            st.session_state.facial_emotion = "Error"
-        st.session_state.last_emotion_time = frame.time
-    return av.VideoFrame.from_ndarray(img, format="bgr24")
-
-if webrtc_ctx.video_processor:
-    webrtc_ctx.video_processor.video_frame_callback = video_frame_callback
-
-# --- Audio Processing ---
-if "audio_buffer" not in st.session_state:
-    st.session_state.audio_buffer = []
-
-def audio_frame_callback(frame: av.AudioFrame):
-    sound = np.frombuffer(frame.to_ndarray(), dtype=np.int16)
-    st.session_state.audio_buffer.append(sound)
-
-webrtc_streamer(
-    key="affectlink-audio",
-    mode=WebRtcMode.RECVONLY,
-    rtc_configuration=RTC_CONFIGURATION, # Use the new config
-    media_stream_constraints={"video": False, "audio": True},
-    audio_frame_callback=audio_frame_callback,
-    async_processing=True,
-)
-
-# --- UI Layout ---
-st.sidebar.header("Facial Emotion")
-st.sidebar.metric("Current Expression", st.session_state.get('facial_emotion', 'N/A'))
-st.sidebar.info("Facial emotion is updated every 5 seconds.")
-st.sidebar.divider()
-
-st.sidebar.header("Audio Analysis")
-is_recording = st.sidebar.checkbox("Start Recording Audio")
-
-st.sidebar.subheader("Transcription:")
-transcription_placeholder = st.sidebar.empty()
-transcription_placeholder.write("_Waiting for audio..._")
-
-st.sidebar.subheader("Text Emotion:")
-text_emotion_placeholder = st.sidebar.empty()
-text_emotion_placeholder.write("_Waiting for audio..._")
-
-st.sidebar.subheader("Speech Emotion:")
-ser_placeholder = st.sidebar.empty()
-ser_placeholder.write("_Waiting for audio..._")
-
-if not is_recording and st.session_state.audio_buffer:
-    audio_data = np.concatenate(st.session_state.audio_buffer)
-    st.session_state.audio_buffer = []
-
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_file:
-        write_wav(tmp_audio_file.name, AUDIO_SAMPLE_RATE, audio_data)
-        with st.spinner("Transcribing audio..."):
-            try:
-                result = whisper_model.transcribe(tmp_audio_file.name, fp16=False)
-                transcribed_text = result['text']
-            except Exception as e:
-                transcribed_text = f"Transcription failed: {e}"
-            transcription_placeholder.write(f'"{transcribed_text}"')
-
-        with st.spinner("Analyzing text emotion..."):
-            if transcribed_text:
-                try:
-                    text_emotions = text_classifier(transcribed_text)[0]
-                    unified_text_scores = {e: 0.0 for e in UNIFIED_EMOTIONS}
-                    for EMO in text_emotions:
-                        unified_emo = TEXT_TO_UNIFIED.get(EMO['label'])
-                        if unified_emo:
-                            unified_text_scores[unified_emo] += EMO['score']
-                    dominant_text_emotion = max(unified_text_scores, key=unified_text_scores.get)
-                except Exception as e:
-                    dominant_text_emotion = f"Text analysis failed: {e}"
-            else:
-                dominant_text_emotion = "No text to analyze."
-            text_emotion_placeholder.write(dominant_text_emotion.capitalize())
-
-        with st.spinner("Analyzing speech emotion..."):
-            try:
-                audio_array, _ = sf.read(tmp_audio_file.name)
-                inputs = ser_feature_extractor(audio_array, sampling_rate=AUDIO_SAMPLE_RATE, return_tensors="pt", padding=True)
-                with torch.no_grad():
-                    logits = ser_model(**inputs).logits
-                scores = torch.nn.functional.softmax(logits, dim=1).squeeze()
-                unified_ser_scores = {e: 0.0 for e in UNIFIED_EMOTIONS}
-                for i, score in enumerate(scores):
-                    raw_emo = ser_model.config.id2label[i]
-                    unified_emo = SER_TO_UNIFIED.get(raw_emo)
-                    if unified_emo:
-                        unified_ser_scores[unified_emo] += score.item()
-                dominant_ser_emotion = max(unified_ser_scores, key=unified_ser_scores.get)
-            except Exception as e:
-                dominant_ser_emotion = f"Speech analysis failed: {e}"
-            ser_placeholder.write(dominant_ser_emotion.capitalize())
-
-        os.unlink(tmp_audio_file.name)
-
-elif is_recording:
-    st.sidebar.warning("Recording audio... Uncheck to stop and process.")