describealign 1.1.2__tar.gz → 1.2.0__tar.gz

{describealign-1.1.2 → describealign-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: describealign
-Version: 1.1.2
+Version: 1.2.0
 Summary: Combines videos with matching audio files (e.g. audio descriptions)
 Author-email: Julian Brown <julbean@proton.me>
 Project-URL: Homepage, https://github.com/julbean/describealign

{describealign-1.1.2 → describealign-1.2.0}/describealign.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: describealign
-Version: 1.1.2
+Version: 1.2.0
 Summary: Combines videos with matching audio files (e.g. audio descriptions)
 Author-email: Julian Brown <julbean@proton.me>
 Project-URL: Homepage, https://github.com/julbean/describealign

{describealign-1.1.2 → describealign-1.2.0}/describealign.py

@@ -461,7 +461,8 @@ def cap_synced_end_points(smooth_path, video_arr, audio_desc_arr):
     smooth_path[-1] = new_end_point
 
 # visualize both the rough and smooth alignments
-def plot_alignment(plot_filename_no_ext, path, smooth_path, quals, runs, bad_clips, ad_timings):
+def plot_alignment(plot_filename_no_ext, path, smooth_path, quals, runs,
+                   bad_clips, ad_timings, similarity_percent):
   scatter_color = [.2,.4,.8]
   lcs_rgba = np.zeros((len(quals),4))
   lcs_rgba[:,:3] = np.array(scatter_color)[None,:]
@@ -500,7 +501,7 @@ def plot_alignment(plot_filename_no_ext, path, smooth_path, quals, runs, bad_cli
       plt.plot(video_times / 60., audio_offsets, 'r-', lw=1, label='Replaced Audio')
   plt.xlabel('Video Time (minutes)')
   plt.ylabel('Audio Description Offset (seconds)')
-  plt.title('Alignment')
+  plt.title(f"Alignment - Media Similarity {similarity_percent:.2f}%")
   plt.legend().legend_handles[0].set_color(scatter_color)
   plt.tight_layout()
   plt.savefig(plot_filename_no_ext + '.png', dpi=400)
@@ -509,6 +510,7 @@ def plot_alignment(plot_filename_no_ext, path, smooth_path, quals, runs, bad_cli
   with open(plot_filename_no_ext + '.txt', 'w') as file:
     rough_clips, median_slope, _ = chunk_path(smooth_path, tol=2e-2)
     video_offset = np.diff(smooth_path[rough_clips[0][0]])[0]
+    print(f"Input file similarity: {similarity_percent:.2f}%", file=file)
     print("Main changes needed to video to align it to audio input:", file=file)
     print(f"Start Offset: {-video_offset:.2f} seconds", file=file)
     print(f"Median Rate Change: {(median_slope-1.)*100:.2f}%", file=file)
@@ -553,6 +555,11 @@ def replace_aligned_segments(video_arr, audio_desc_arr, smooth_path, runs, no_pi
     segment = np.hstack(segment)
     return segment
   
+  # compress dual channel audio to mono for use in pitch corrected stretching
+  # pytsmod's wsola treats channels separately, so without this it sounds weird
+  if not no_pitch_correction:
+    audio_desc_arr_mono = np.mean(audio_desc_arr, axis=0)
+  
   x,y = zip(*smooth_path)
   for run in runs:
     run_length_seconds = y[run[-1][1]] - y[run[0][0]]
@@ -573,7 +580,7 @@ def replace_aligned_segments(video_arr, audio_desc_arr, smooth_path, runs, no_pi
         anchor_point_pair[1][-1] -= 1
         anchor_y_offset = anchor_point_pair[1][0]
         anchor_point_pair[1,:] -= anchor_y_offset
-        stretched_audio = pytsmod.wsola(audio_desc_arr, anchor_point_pair)
+        stretched_audio = pytsmod.wsola(audio_desc_arr_mono, anchor_point_pair)
       video_arr[:,slice(*anchor_points[1,clip_index:clip_index+2])] = stretched_audio
 
 # identify which segments of the replaced audio actually have the describer speaking
@@ -755,7 +762,7 @@ def get_ffprobe():
 def get_closest_key_frame_time(video_file, time):
   if time <= 0:
     return 0
-  key_frames = ffmpeg.probe(video_file, cmd=get_ffprobe(), select_streams='v',
+  key_frames = ffmpeg.probe(video_file, cmd=get_ffprobe(), select_streams='V',
                             show_frames=None, skip_frame='nokey')['frames']
   key_frame_times = np.array([float(frame['pts_time']) for frame in key_frames] + [0])
   return np.max(key_frame_times[key_frame_times <= time])
@@ -891,6 +898,11 @@ def combine(video, audio, smoothness=50, stretch_audio=False, keep_non_ad=False,
       display("   output file already exists, skipping...", display_func)
       continue
     
+    # print warning if output file's full path is longer than Windows MAX_PATH (260)
+    full_output_filename = os.path.abspath(output_filename)
+    if IS_RUNNING_WINDOWS and len(full_output_filename) >= 260:
+      display("   WARNING: very long output path, ffmpeg may fail...", display_func)
+    
     video_arr = parse_audio_from_file(video_file)
     audio_desc_arr = parse_audio_from_file(audio_desc_file)
     video_spec_raw, video_timings = tokenize_audio(video_arr)
@@ -903,6 +915,13 @@ def combine(video, audio, smoothness=50, stretch_audio=False, keep_non_ad=False,
     
     path, quals = rough_align(video_spec, audio_desc_spec, video_timings, audio_desc_timings)
     
+    similarity_ratio = float(len(quals)) / max(video_spec.shape[0], audio_desc_spec.shape[0])
+    similarity_percent = min(100, 100 * similarity_ratio)
+    if similarity_percent < 10:
+      display(f"   WARNING: similarity {similarity_percent:.1f}%, likely mismatched files", display_func)
+    if similarity_percent > 90:
+      display(f"   WARNING: similarity {similarity_percent:.1f}%, likely undescribed media", display_func)
+    
     smooth_path, runs, bad_clips, clips = smooth_align(path, quals, smoothness)
     
     cap_synced_end_points(smooth_path, video_arr, audio_desc_arr)
@@ -951,7 +970,8 @@ def combine(video, audio, smoothness=50, stretch_audio=False, keep_non_ad=False,
     del video_arr
     if PLOT_ALIGNMENT_TO_FILE:
       plot_filename_no_ext = os.path.join(alignment_dir, os.path.splitext(os.path.split(video_file)[1])[0])
-      plot_alignment(plot_filename_no_ext, path, smooth_path, quals, runs, bad_clips, ad_timings)
+      plot_alignment(plot_filename_no_ext, path, smooth_path, quals,
+                     runs, bad_clips, ad_timings, similarity_percent)
   display("All files processed.", display_func)
 
 def write_config_file(config_path, settings):
@@ -1206,6 +1226,7 @@ def main_gui():
         window.disable()
         sg.Popup('Error: empty input field.', font=('Arial', 20))
         window.enable()
+        window['-VIDEO_FILES-'].set_focus()
         continue
       video_files = values['-VIDEO_FILES-'].split(';')
       if len(video_files) == 1:
@@ -1213,11 +1234,15 @@ def main_gui():
       audio_files = values['-AUDIO_FILES-'].split(';')
       if len(audio_files) == 1:
         audio_files = audio_files[0]
+      window.disable()
       combine_gui(video_files, audio_files, config_path)
+      window.enable()
+      window['-VIDEO_FILES-'].set_focus()
     if event == 'Settings':
       window.disable()
       settings_gui(config_path)
       window.enable()
+      window['-VIDEO_FILES-'].set_focus()
     if event == sg.WIN_CLOSED:
       break
   window.close()

describealign-1.2.0/version

@@ -0,0 +1 @@
+1.2.0

describealign-1.1.2/version

@@ -1 +0,0 @@
-1.1.2