whisper_edit.py

@@ -50,8 +50,8 @@ class Commands:
     'drop_segment': drop_segment,
     'chapter_break': chapter_break,
   }
-input_file = 'input.mp4'
-output_file = 'output.mp4'
+input_file = sys.argv[1]
+output_file = sys.argv[2]
 def speech_to_text():
   model = whisper.load_model(MODEL)
   modify_model(model)
@@ -116,7 +116,8 @@ def find_commands(timestamps):
 
 # Hack. When dropping a segment I see a split-second of the end of
 # the dropped segment. So add a small fuzz factor for that.
-DROP_SEGMENT_DELTA = 0.2
+DROP_SEGMENT_DELTA = 0.0
+KEEP_SEGMENT_DELTA = 0.8
 
 # Returns list of (start, end) tuples of intervals to keep.
 def intervals_to_keep(commands):
@@ -127,7 +128,7 @@ def intervals_to_keep(commands):
     match cmd:
       case Commands.keep_segment:
         # Keep until the start of the command.
-        keep_end = begin_ts
+        keep_end = begin_ts - KEEP_SEGMENT_DELTA
         keep_intervals.append([keep_start, keep_end])
         # Next (possibly) starts at end of command.
         keep_start = end_ts
@@ -187,9 +188,9 @@ def main():
   sts = speech_to_text()
   word_ts = word_level_timestamps(sts)
   commands = find_commands(word_ts)
-  print(commands)
+  print(f'Commands: {commands}')
   keep_intervals = intervals_to_keep(commands)
-  print(keep_intervals)
+  print(f'Keeping intervals: {keep_intervals}')
 
   vid = VideoFileClip(input_file)
 
@@ -202,6 +203,7 @@ def main():
 
   print("\n\n\n----- Writing out edited video... -----")
   no_dead_air_video.write_videofile(output_file,
+  #edited_vid.write_videofile(output_file,
         #fps=60,
         preset='ultrafast',
         codec='libx264',