When the language is set to Chinese,

the method of converting simplified Chinese to traditional Chinese is changed to using the method provided by zhconv.

When the "--merge_subtitle_with_sources" argument is enabled,
the video file extension will be appended with the subtitle language information, such as .en, .zh, .jp, etc.

The downloaded YouTube videos will be added with the format [vcodec^=avc1].

app.py

@@ -19,6 +19,7 @@ from src.hooks.subTaskProgressListener import SubTaskProgressListener
 from src.hooks.whisperProgressHook import create_progress_listener_handle
 from src.languages import _TO_LANGUAGE_CODE
 from src.languages import get_language_names
+from src.languages import get_language_from_name
 from src.modelCache import ModelCache
 from src.prompts.jsonPromptStrategy import JsonPromptStrategy
 from src.prompts.prependPromptStrategy import PrependPromptStrategy
@@ -38,6 +39,7 @@ from src.whisper.abstractWhisperContainer import AbstractWhisperContainer
 from src.whisper.whisperFactory import create_whisper_container
 
 import shutil
+import zhconv
 
 # Configure more application defaults in config.json5
 
@@ -102,14 +104,11 @@ class WhisperTranscriber:
                                          vad, vadMergeWindow, vadMaxMergeSize, 
                                          word_timestamps: bool = False, highlight_words: bool = False, 
                                          progress=gr.Progress()):
-        decodeOptions = dict(word_timestamps=word_timestamps)
-        if languageName == "Chinese":
-            decodeOptions.update(initial_prompt="繁體: ")
-            self.app_config.vad_initial_prompt_mode = "prepend_all_segments"
 
         vadOptions = VadOptions(vad, vadMergeWindow, vadMaxMergeSize, self.app_config.vad_padding, self.app_config.vad_prompt_window, self.app_config.vad_initial_prompt_mode)
 
-        return self.transcribe_webui(modelName, languageName, urlData, multipleFiles, microphoneData, task, vadOptions, highlight_words=highlight_words, progress=progress, **decodeOptions)
+        return self.transcribe_webui(modelName, languageName, urlData, multipleFiles, microphoneData, task, vadOptions, 
+                                     word_timestamps=word_timestamps, highlight_words=highlight_words, progress=progress)
 
     # Entry function for the full tab
     def transcribe_webui_full(self, modelName, languageName, urlData, multipleFiles, microphoneData, task, 
@@ -143,10 +142,6 @@ class WhisperTranscriber:
         else:
             temperature = [temperature]
 
-        if languageName == "Chinese":
-            initial_prompt = "繁體: " + initial_prompt
-            self.app_config.vad_initial_prompt_mode = "prepend_all_segments"
-
         vadOptions = VadOptions(vad, vadMergeWindow, vadMaxMergeSize, vadPadding, vadPromptWindow, vadInitialPromptMode)
 
         return self.transcribe_webui(modelName, languageName, urlData, multipleFiles, microphoneData, task, vadOptions,
@@ -163,7 +158,8 @@ class WhisperTranscriber:
             sources = self.__get_source(urlData, multipleFiles, microphoneData)
 
             try:
-                selectedLanguage = languageName.lower() if len(languageName) > 0 else None
+                langObj = get_language_from_name(languageName)
+                selectedLanguage = languageName.lower() if languageName is not None and len(languageName) > 0 else None
                 selectedModel = modelName if modelName is not None else "base"
 
                 model = create_whisper_container(whisper_implementation=self.app_config.whisper_implementation, 
@@ -266,13 +262,14 @@ class WhisperTranscriber:
                                 srt_path = source_download[0]
                                 save_path = os.path.join(self.app_config.output_dir, source.source_name)
                                 save_without_ext, ext = os.path.splitext(save_path)
-                                output_with_srt = save_without_ext + ".srt" + ext
+                                lang_ext = "." + langObj.code if langObj is not None else ""
+                                output_with_srt = save_without_ext + lang_ext + ext
                                 
                                 #ffmpeg -i "input.mp4" -i "input.srt" -c copy -c:s mov_text output.mp4
                                 input_file = ffmpeg.input(source.source_path)
                                 input_srt = ffmpeg.input(srt_path)
                                 out = ffmpeg.output(input_file, input_srt, output_with_srt, vcodec='copy', acodec='copy', scodec='mov_text')
-                                outRsult = out.run()
+                                outRsult = out.run(overwrite_output=True)
                             except Exception as e:
                                 # Ignore error - it's just a cleanup
                                 print("Error merge subtitle with source file: \n" + source.source_path + ", \n" + str(e), outRsult)
@@ -439,6 +436,12 @@ class WhisperTranscriber:
         srt = self.__get_subs(result["segments"], "srt", languageMaxLineWidth, highlight_words=highlight_words)
         json_result = json.dumps(result, indent=4, ensure_ascii=False)
 
+        if language == "zh":
+            vtt = zhconv.convert(vtt, "zh-tw")
+            srt = zhconv.convert(srt, "zh-tw")
+            text = zhconv.convert(text, "zh-tw")
+            json_result = zhconv.convert(json_result, "zh-tw")
+
         output_files = []
         output_files.append(self.__create_file(srt, output_dir, source_name + "-subs.srt"));
         output_files.append(self.__create_file(vtt, output_dir, source_name + "-subs.vtt"));

requirements-fasterWhisper.txt

@@ -6,4 +6,5 @@ yt-dlp
 json5
 torch
 torchaudio
-more_itertools
+more_itertools
+zhconv

requirements-whisper.txt

@@ -6,4 +6,5 @@ gradio==3.36.0
 yt-dlp
 torchaudio
 altair
-json5
+json5
+zhconv

requirements.txt

@@ -6,4 +6,5 @@ yt-dlp
 json5
 torch
 torchaudio
-more_itertools
+more_itertools
+zhconv

src/download.py

@@ -29,7 +29,7 @@ def _perform_download(url: str, maxDuration: int = None, outputTemplate: str = N
         destinationDirectory = mkdtemp()
 
     ydl_opts = {
-        "format": "bestaudio/best" if onlyAudio else "bestvideo[ext=mp4]+bestaudio[ext=m4a]/best",
+        "format": "bestaudio/best" if onlyAudio else "bestvideo[ext=mp4][vcodec^=avc1]+bestaudio[ext=m4a]/best",
         'paths': {
             'home': destinationDirectory
         }

src/vad.py

@@ -204,7 +204,7 @@ class AbstractTranscription(ABC):
                 detected_language = languageCounter.most_common(1)[0][0] if len(languageCounter) > 0 else None
 
                 print("Running whisper from ", format_timestamp(segment_start), " to ", format_timestamp(segment_end), ", duration: ", 
-                    segment_duration, "expanded: ", segment_expand_amount, "prompt: ", segment_prompt, "language: ", detected_language)
+                    segment_duration, "expanded: ", segment_expand_amount, ", prompt: ", segment_prompt, ", detected language: ", detected_language)
 
                 perf_start_time = time.perf_counter()