Update pipeline.py

pipeline.py

@@ -287,16 +287,21 @@ def draw_metric_bars(panel, bars_x_start, bar_y_top, bar_height, bar_width,
 
 
 def draw_question_overlay(frame, text: str, width: int, height: int):
-    """Draws a semi-transparent wrapped-text box above the wheel and bars."""
+    """
+    Draws a semi-transparent wrapped-text box above the wheel and bars.
+
+    CHANGE: font_scale reduced from 0.65 → 0.52 and line_height from 35 → 28
+    so the question text appears smaller on screen.
+    """
     if frame is None or not text:
         return frame
     try:
-        font        = cv2.FONT_HERSHEY_DUPLEX
-        font_scale  = 0.65
-        thickness   = 1
-        side_margin = 50
+        font         = cv2.FONT_HERSHEY_DUPLEX
+        font_scale   = 0.52          # CHANGED: was 0.65 — smaller text on screen
+        thickness    = 1
+        side_margin  = 50
         bottom_limit = height - 270
-        line_height  = 35
+        line_height  = 28            # CHANGED: was 35 — tighter rows to match smaller font
 
         max_w = width - (2 * side_margin)
         words, lines, current_line = text.split(' '), [], ""
@@ -644,6 +649,10 @@ def process_full_video(video_path: str, output_dir: str, questions_config: list,
     Annotate the video with emotion wheel, face box, and live metric bars.
     Returns the path to the raw annotated video (no audio).
     Returns None on fatal failure.
+
+    CHANGE: Metric bars (confidence, clarity, stress) are forced to 0.0 whenever
+    active_answer is None — i.e. during AI speech / between questions — so that
+    the AI's own voice does not pollute the live analysis display.
     """
     if not os.path.exists(video_path):
         log.error("process_full_video: video not found – %s", video_path)
@@ -721,6 +730,9 @@ def process_full_video(video_path: str, output_dir: str, questions_config: list,
                 next_text = f"Q) {next_q['question_text']}" if next_q else "Preparing..."
 
                 # ── Live audio (every 10 frames) ─────────────────────────────
+                # We still compute live_scores every 10 frames so they are
+                # ready to display the moment the candidate starts answering.
+                # The decision of whether to SHOW them is made at render time.
                 if frame_idx % 10 == 0 and len(full_audio) > 1:
                     try:
                         start_s = max(0, int((current_time - 3) * sr))
@@ -773,11 +785,25 @@ def process_full_video(video_path: str, output_dir: str, questions_config: list,
                     draw_full_emotion_wheel(
                         frame, (width - 130, height - 100), 90, smooth_v, smooth_a, dom_emo
                     )
+
+                    # CHANGE: only show real bar values while the candidate is
+                    # answering (active_answer is not None).  During AI speech
+                    # or between questions the bars are rendered as 0 so the
+                    # AI's own voice doesn't pollute the display.
+                    if active_answer:
+                        bar_confidence = live_scores["confidence_audio"]
+                        bar_clarity    = live_scores["clarity"]
+                        bar_stress     = live_scores["stress"]
+                    else:
+                        bar_confidence = 0.0
+                        bar_clarity    = 0.0
+                        bar_stress     = 0.0
+
                     draw_metric_bars(
                         frame, 30, height - 160, 28, 200, 6,
-                        live_scores["confidence_audio"],
-                        live_scores["clarity"],
-                        live_scores["stress"],
+                        bar_confidence,
+                        bar_clarity,
+                        bar_stress,
                     )
 
                     if not active_answer: