diff --git a/aider/website/_includes/recording.js b/aider/website/_includes/recording.js
index d6ee290f9..f762582b0 100644
--- a/aider/website/_includes/recording.js
+++ b/aider/website/_includes/recording.js
@@ -180,6 +180,20 @@ document.addEventListener('DOMContentLoaded', function() {
     }, 3000);
   }
   
+  // Function to use browser's TTS as fallback
+  function useBrowserTTS(text) {
+    if ('speechSynthesis' in window) {
+      console.log('Using browser TTS fallback');
+      const utterance = new SpeechSynthesisUtterance(text);
+      utterance.rate = 1.0;
+      utterance.pitch = 1.0;
+      utterance.volume = 1.0;
+      window.speechSynthesis.speak(utterance);
+      return true;
+    }
+    return false;
+  }
+  
   // Function to play pre-generated TTS audio files
   function speakText(text, timeInSeconds) {
     // Format time for filename (MM-SS)
@@ -200,19 +214,14 @@ document.addEventListener('DOMContentLoaded', function() {
     // Error handling with fallback to browser TTS
     audio.onerror = () => {
       console.warn(`Failed to load audio: ${audioPath}`);
-      // Fallback to browser TTS
-      if ('speechSynthesis' in window) {
-        const utterance = new SpeechSynthesisUtterance(text);
-        utterance.rate = 1.0;
-        utterance.pitch = 1.0;
-        utterance.volume = 1.0;
-        window.speechSynthesis.speak(utterance);
-      }
+      useBrowserTTS(text);
     };
     
     // Play the audio
     audio.play().catch(e => {
       console.warn(`Error playing audio: ${e.message}`);
+      // Also fall back to browser TTS if play() fails
+      useBrowserTTS(text);
     });
   }
   
diff --git a/scripts/recording_audio.py b/scripts/recording_audio.py
index eed084e87..b854f4f5c 100755
--- a/scripts/recording_audio.py
+++ b/scripts/recording_audio.py
@@ -5,6 +5,7 @@ Usage: python scripts/recording_audio.py path/to/recording.md
 """
 
 import argparse
+import json
 import os
 import re
 import sys
@@ -80,6 +81,32 @@ def generate_audio_openai(text, output_file, voice=VOICE):
         return False
 
 
+def load_metadata(output_dir):
+    """Load the audio metadata JSON file if it exists."""
+    metadata_file = os.path.join(output_dir, "metadata.json")
+    
+    if os.path.exists(metadata_file):
+        try:
+            with open(metadata_file, "r") as f:
+                return json.load(f)
+        except json.JSONDecodeError:
+            print(f"Warning: Could not parse metadata file {metadata_file}, will recreate it")
+    
+    return {}
+
+def save_metadata(output_dir, metadata):
+    """Save the audio metadata to JSON file."""
+    metadata_file = os.path.join(output_dir, "metadata.json")
+    
+    with open(metadata_file, "w") as f:
+        json.dump(metadata, f, indent=2)
+
+def get_timestamp_key(time_sec):
+    """Generate a consistent timestamp key format for metadata."""
+    minutes = time_sec // 60
+    seconds = time_sec % 60
+    return f"{minutes:02d}-{seconds:02d}"
+
 def main():
     parser = argparse.ArgumentParser(description="Generate TTS audio for recording commentary.")
     parser.add_argument("markdown_file", help="Path to the recording markdown file")
@@ -90,6 +117,9 @@ def main():
     parser.add_argument(
         "--dry-run", action="store_true", help="Print what would be done without generating audio"
     )
+    parser.add_argument(
+        "--force", action="store_true", help="Force regeneration of all audio files"
+    )
 
     args = parser.parse_args()
 
@@ -113,15 +143,52 @@ def main():
 
     print(f"Found {len(markers)} commentary markers")
 
+    # Load existing metadata
+    metadata = load_metadata(output_dir)
+    
+    # Create a dictionary of current markers for easier comparison
+    current_markers = {}
+    for time_sec, message in markers:
+        timestamp_key = get_timestamp_key(time_sec)
+        current_markers[timestamp_key] = message
+    
+    # Track files that need to be deleted (no longer in the markdown)
+    files_to_delete = []
+    for timestamp_key in metadata:
+        if timestamp_key not in current_markers:
+            files_to_delete.append(f"{timestamp_key}.mp3")
+    
+    # Delete files that are no longer needed
+    if files_to_delete and not args.dry_run:
+        for filename in files_to_delete:
+            file_path = os.path.join(output_dir, filename)
+            if os.path.exists(file_path):
+                print(f"Removing obsolete file: {filename}")
+                os.remove(file_path)
+    elif files_to_delete:
+        print(f"Would remove {len(files_to_delete)} obsolete files: {', '.join(files_to_delete)}")
+
     # Generate audio for each marker
     for time_sec, message in markers:
+        timestamp_key = get_timestamp_key(time_sec)
+        filename = f"{timestamp_key}.mp3"
+        output_file = os.path.join(output_dir, filename)
+        
+        # Check if we need to generate this file
+        needs_update = args.force or (
+            timestamp_key not in metadata or 
+            metadata[timestamp_key] != message
+        )
+        
         minutes = time_sec // 60
         seconds = time_sec % 60
-        timestamp = f"{minutes:02d}-{seconds:02d}"
-        filename = f"{timestamp}.mp3"
-        output_file = os.path.join(output_dir, filename)
-
+        
         print(f"Marker at {minutes}:{seconds:02d} - {message}")
+        
+        if not needs_update:
+            print(f"  ✓ Audio file already exists with correct content")
+            continue
+        
         if args.dry_run:
             print(f"  Would generate: {output_file}")
         else:
@@ -129,8 +196,19 @@ def main():
             success = generate_audio_openai(message, output_file, voice=selected_voice)
             if success:
                 print(f"  ✓ Generated audio file")
+                # Update metadata with the new message
+                metadata[timestamp_key] = message
             else:
                 print(f"  ✗ Failed to generate audio")
+    
+    # Save updated metadata
+    if not args.dry_run:
+        # Remove entries for deleted files
+        for timestamp_key in list(metadata.keys()):
+            if timestamp_key not in current_markers:
+                del metadata[timestamp_key]
+        
+        save_metadata(output_dir, metadata)
 
 
 if __name__ == "__main__":