describealign 2.0.1__tar.gz → 2.0.3__tar.gz

{describealign-2.0.1 → describealign-2.0.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: describealign
-Version: 2.0.1
+Version: 2.0.3
 Summary: Combines videos with matching audio files (e.g. audio descriptions)
 Author-email: Julian Brown <julbean@proton.me>
 License-Expression: GPL-3.0-only
@@ -13,7 +13,7 @@ Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: ffmpeg_python~=0.2.0
-Requires-Dist: static-ffmpeg~=2.5
+Requires-Dist: static-ffmpeg~=3.0
 Requires-Dist: matplotlib~=3.9
 Requires-Dist: numpy<3.0,>=1.21
 Requires-Dist: scipy~=1.10

{describealign-2.0.1 → describealign-2.0.3}/describealign.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: describealign
-Version: 2.0.1
+Version: 2.0.3
 Summary: Combines videos with matching audio files (e.g. audio descriptions)
 Author-email: Julian Brown <julbean@proton.me>
 License-Expression: GPL-3.0-only
@@ -13,7 +13,7 @@ Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: ffmpeg_python~=0.2.0
-Requires-Dist: static-ffmpeg~=2.5
+Requires-Dist: static-ffmpeg~=3.0
 Requires-Dist: matplotlib~=3.9
 Requires-Dist: numpy<3.0,>=1.21
 Requires-Dist: scipy~=1.10

{describealign-2.0.1 → describealign-2.0.3}/describealign.egg-info/requires.txt

@@ -1,5 +1,5 @@
 ffmpeg_python~=0.2.0
-static-ffmpeg~=2.5
+static-ffmpeg~=3.0
 matplotlib~=3.9
 numpy<3.0,>=1.21
 scipy~=1.10

{describealign-2.0.1 → describealign-2.0.3}/describealign.py

@@ -1,4 +1,4 @@
-__version__ = '2.0.1'
+__version__ = '2.0.3'
 
 # combines videos with matching audio files (e.g. audio descriptions)
 # input: video or folder of videos and an audio file or folder of audio files
@@ -141,6 +141,11 @@ def run_async_ffmpeg_command(command, media_arr, err_msg):
     print(e.stderr.decode('utf-8'))
     raise
 
+def get_ffmpeg_version():
+  ffmpeg_command = ffmpeg.input('').output('', version='')
+  stdout, _ = run_ffmpeg_command(ffmpeg_command, "get version information")
+  return float(str(stdout).split('version ')[1][:2])
+
 # read audio from file with ffmpeg and convert to numpy array
 def parse_audio_from_file(media_file, num_channels=2):
   # retrieve only the first audio track, injecting silence/trimming to force timestamps to match up
@@ -149,7 +154,6 @@ def parse_audio_from_file(media_file, num_channels=2):
                                                    af='aresample=async=1:first_pts=0', map='0:a:0',
                                                    ac=num_channels, ar=AUDIO_SAMPLE_RATE, loglevel='error')
   media_stream, _ = run_ffmpeg_command(ffmpeg_command, f"parse audio from input file: {media_file}")
-  # media_arr = np.frombuffer(media_stream, np.int16).astype(np.float32).reshape((-1, num_channels)).T
   media_arr = np.frombuffer(media_stream, np.int16).astype(np.float16).reshape((-1, num_channels)).T
   return media_arr
 
@@ -452,7 +456,8 @@ def get_closest_key_frame_time(video_file, time):
 
 # outputs a new media file with the replaced audio (which includes audio descriptions)
 def write_replaced_media_to_disk(output_filename, media_arr, video_file=None, audio_desc_file=None,
-                                 setts_cmd=None, video_offset=None, after_start_key_frame=None):
+                                 setts_cmd=None, video_offset=None, after_start_key_frame=None,
+                                 median_slope=1.):
   # if a media array is given, stretch_audio is enabled and media_arr should be added to the video
   if media_arr is not None:
     media_input = ffmpeg.input('pipe:', format='s16le', acodec='pcm_s16le', ac=2, ar=AUDIO_SAMPLE_RATE)
@@ -481,6 +486,8 @@ def write_replaced_media_to_disk(output_filename, media_arr, video_file=None, au
     audio_codec = 'copy' if os.path.splitext(audio_desc_file)[1] != '.wav' else 'aac'
     # flac audio may only have experimental support in some video containers (e.g. mp4)
     standards = 'normal' if os.path.splitext(audio_desc_file)[1] != '.flac' else 'experimental'
+    # stretch subtitle durations along with video so they don't overlap or have gaps
+    sub_stretch = f':duration=\'DURATION*{1./median_slope:.6f}\''
     # add frag_keyframe flag to prevent some players from ignoring audio/video start offsets
     # set both pts and dts simultaneously in video manually, as ts= does not do the same thing
     write_command = ffmpeg.output(media_input, original_video, output_filename,
@@ -488,16 +495,30 @@ def write_replaced_media_to_disk(output_filename, media_arr, video_file=None, au
                                   max_interleave_delta='0', loglevel='error',
                                   strict=standards, movflags='frag_keyframe',
                                   **{'bsf:v': f'setts=pts=\'{setts_cmd}\':dts=\'{setts_cmd}\'',
-                                     'bsf:s': f'setts=ts=\'{setts_cmd}\'',
+                                     'bsf:s': f'setts=ts=\'{setts_cmd}\'' + sub_stretch,
                                      "disposition:a:0": "default+visual_impaired+descriptions",
                                      "metadata:s:a:0": "title=AD"}).overwrite_output()
     run_ffmpeg_command(write_command, f"write output file: {output_filename}")
 
+def get_static_ffmpeg_version():
+  # if running from compiled binary, assume correct version of static_ffmpeg
+  if "__compiled__" in globals() or getattr(sys, 'frozen', False):
+    return 3
+  import importlib
+  static_ffmpeg_version = importlib.metadata.version('static_ffmpeg')
+  return float(static_ffmpeg_version[:2])
+
 # check whether static_ffmpeg has already installed ffmpeg and ffprobe
 def is_ffmpeg_installed():
   ffmpeg_dir = static_ffmpeg.run.get_platform_dir()
   indicator_file = os.path.join(ffmpeg_dir, "installed.crumb")
-  return os.path.exists(indicator_file)
+  if not os.path.exists(indicator_file):
+    return False
+  if get_ffmpeg_version() < 6:
+    print("Old ffmpeg version detected, updating to newer version...")
+    os.remove(indicator_file)
+    return False
+  return True
 
 def get_energy(arr):
   # downsample of 105, hann size 15, downsample by 2 gives 210 samples per second, ~65 halfwindows/second
@@ -599,6 +620,9 @@ def align(video_features, audio_desc_features, video_energy, audio_desc_energy):
     bins = np.floor(bins).astype(int)
     np.clip(bins, 0, 6, out=bins)
     audio_desc_bins.append(np.dot(bins, 7**np.arange(num_bins)).tolist())
+  del feature
+  del norm
+  del bins
   
   def pairwise_intersection(set1, set2, set3):
     return (set1 & set2).union((set1 & set3), (set2 & set3))
@@ -635,6 +659,12 @@ def align(video_features, audio_desc_features, video_energy, audio_desc_energy):
       best_so_far.add((video_index, audio_desc_index, cum_qual))
       backpointers[(video_index, audio_desc_index)] = (prev_video_index, prev_audio_desc_index)
   del video_dicts
+  del video_dict
+  del audio_desc_bins
+  del video_features_mean_sub
+  del audio_desc_features_mean_sub
+  del video_uniform_norms
+  del audio_desc_uniform_norms
   path = [best_so_far[-1][:2]]
   while path[-1][:2] in backpointers:
     # failsafe to prevent an infinite loop that should never happen anyways
@@ -887,6 +917,7 @@ def align(video_features, audio_desc_features, video_energy, audio_desc_energy):
         seen_points.add(point)
         points[i].append((j, cluster_index, qual))
   del seen_points
+  del video_interp
   points = [sorted(point) for point in points]
   
   best_so_far = SortedList(key=lambda x:x[0])
@@ -897,6 +928,7 @@ def align(video_features, audio_desc_features, video_energy, audio_desc_energy):
   prev_cache[0] = (0, 0, -1, 0, 0)  # video_index, audio_desc_index, cluster_index, qual, cum_qual
   reversed_min_points = [min(x)[0] if len(x) > 0 else np.inf for x in points[::-1]]
   forward_min = list(itertools.accumulate(reversed_min_points, min))[::-1]
+  del reversed_min_points
   for i in range(len(audio_desc_features_scaled)):
     for j, cluster_index, qual in points[i]:
       cur_index = best_so_far.bisect_right((j,))
@@ -926,6 +958,7 @@ def align(video_features, audio_desc_features, video_energy, audio_desc_energy):
       if cluster_last[3] < cum_qual_cluster_jump:
         clusters_best_so_far[cluster_index] = (j, i, qual, cum_qual_cluster_jump)
       backpointers[(j, i)] = (prev_j, prev_i, prev_cluster_index, prev_qual, best_prev_cum_qual)
+  
   path = [best_so_far[-1]]
   while path[-1][:2] in backpointers:
     path.append(backpointers[path[-1][:2]])
@@ -1005,6 +1038,9 @@ def combine(video, audio, stretch_audio=False, yes=False, prepend="ad_", no_pitc
   
   # if ffmpeg isn't installed, install it
   if not is_ffmpeg_installed():
+    if get_static_ffmpeg_version() < 3:
+      print(f"  ERROR: outdated static_ffmpeg version")
+      raise ImportError("static_ffmpeg must be at least version 3.0")
     print("Downloading and installing ffmpeg (media editor, 50 MB download)...")
     get_ffmpeg()
     if not is_ffmpeg_installed():
@@ -1084,7 +1120,8 @@ def combine(video, audio, stretch_audio=False, yes=False, prepend="ad_", no_pitc
       video_arr *= (2**15 - 2.) / np.max(np.abs(video_arr))
       
       print("  processing output file...                   \r", end='')
-      write_replaced_media_to_disk(output_filename, video_arr, None if has_audio_extension else video_file)
+      write_replaced_media_to_disk(output_filename, video_arr, None if has_audio_extension else video_file,
+                                   median_slope=median_slope)
       del video_arr
     else:
       video_offset = video_times[0] - audio_desc_times[0]
@@ -1093,7 +1130,8 @@ def combine(video, audio, stretch_audio=False, yes=False, prepend="ad_", no_pitc
       print("  processing output file...                   \r", end='')
       setts_cmd = encode_fit_as_ffmpeg_expr(audio_desc_times, video_times, video_offset)
       write_replaced_media_to_disk(output_filename, None, video_file, audio_desc_file,
-                                   setts_cmd, video_offset, after_start_key_frame)
+                                   setts_cmd, video_offset, after_start_key_frame,
+                                   median_slope=median_slope)
     
     if PLOT_ALIGNMENT_TO_FILE:
       plot_filename_no_ext = os.path.join(alignment_dir, os.path.splitext(os.path.split(video_file)[1])[0])
@@ -1132,7 +1170,7 @@ if wx is not None:
   
   class DialogSettings(wx.Dialog):
     def __init__(self, parent, config_path, is_dark):
-      wx.Dialog.__init__(self, parent, title="Settings - describealign", size=wx.Size(450,330),
+      wx.Dialog.__init__(self, parent, title="Settings - describealign", size=wx.Size(450,370),
                          style=wx.DEFAULT_DIALOG_STYLE|wx.TAB_TRAVERSAL)
       # setting the GUI dialog's font causes all contained elements to inherit that font by default
       self.SetFont(wx.Font(*gui_font))
@@ -1209,14 +1247,14 @@ if wx is not None:
       #
       self.SetSizer(sizer_dialog)
       sizer_dialog.Add(self.text_header, 0, wx.ALL, 5)
-      sizer_dialog.Add(sizer_output_dir, 1, wx.LEFT|wx.RIGHT|wx.EXPAND, 2)
-      sizer_dialog.Add(sizer_alignment_dir, 1, wx.LEFT|wx.RIGHT|wx.EXPAND, 2)
-      sizer_dialog.Add(sizer_prepend, 1, wx.LEFT|wx.EXPAND, 5)
-      sizer_dialog.Add(sizer_stretch_audio_no_pitch_correction_outer, 1, wx.LEFT|wx.EXPAND, 5)
+      sizer_dialog.Add(sizer_output_dir, 3, wx.LEFT|wx.RIGHT|wx.EXPAND, 2)
+      sizer_dialog.Add(sizer_alignment_dir, 3, wx.LEFT|wx.RIGHT|wx.EXPAND, 2)
+      sizer_dialog.Add(sizer_prepend, 3, wx.LEFT|wx.EXPAND, 5)
+      sizer_dialog.Add(sizer_stretch_audio_no_pitch_correction_outer, 3, wx.LEFT|wx.EXPAND, 5)
       sizer_stretch_audio_no_pitch_correction_outer.Add(panel_stretch_audio_no_pitch_correction,
                                                         1, wx.LEFT|wx.EXPAND, 5)
       sizer_stretch_audio_no_pitch_correction_outer.Add((0, 0), 2, wx.EXPAND, 5)  # spacer
-      sizer_dialog.Add(sizer_save_cancel, 2, wx.BOTTOM|wx.EXPAND, 5)
+      sizer_dialog.Add(sizer_save_cancel, 5, wx.BOTTOM|wx.EXPAND, 5)
       sizer_prepend.Add(self.text_prepend, 0, wx.ALL|wx.ALIGN_CENTER_VERTICAL, 5)
       sizer_prepend.Add(self.text_ctrl_prepend, 0, wx.ALIGN_CENTER_VERTICAL, 5)
       sizer_output_dir.Add(self.static_box_sizer_output, 1, wx.LEFT|wx.RIGHT|wx.ALIGN_CENTER_VERTICAL, 5)

{describealign-2.0.1 → describealign-2.0.3}/requirements.txt

@@ -1,5 +1,5 @@
 ffmpeg_python~=0.2.0
-static-ffmpeg~=2.5
+static-ffmpeg~=3.0
 matplotlib~=3.9
 numpy>=1.21,<3.0
 scipy~=1.10