pyVideoTrans

Appearance

Explanation of Advanced Settings Options ​

You can customize various parameters in the top menu under Tools/Options -> Advanced Options for finer control. See the image below.

【Common】 ​

  • Software interface language: Set the software's interface language. Requires a restart to take effect.

  • Countdown for single video translation pause: Countdown in seconds for a single video translation.

  • Set output directory: Directory to save results (video separation, subtitles, dubbing). Defaults to the 'output' folder.

  • LLM re-segmentation How many subtitles are sent each time: When re-segmenting sentences in the LLM large model, the number of subtitles sent each time is important. A larger value results in better sentence segmentation. Sending all subtitles at once is optimal, but this is limited by the maximum output token and context (max_token). An excessively long input may exceed the AI limit and fail. The default is 20 subtitles.

  • AI provider for LLM re-segmentation: AI provider for LLM re-segmentation, supports 'openai' or 'deepseek'.

  • Gemini speech recognition batch slice count: Number of audio slices per request for Gemini recognition. Larger values improve accuracy but increase failure rate.

  • Disable desktop notifications: Disable desktop notifications for task completion or failure.

  • Force serial processing for batch translation: Process batch translation sequentially (one by one) instead of in parallel.

  • Show all parameters?: To avoid confusion caused by too many parameters, most parameters are hidden by default on the main interface. Selecting this option will switch to displaying all parameters by default.

  • Process Maximum: Process Maximum

  • Number of GPU tasks[restart]: The number of GPU tasks that can be executed simultaneously should be set to 1 unless the video memory is greater than 20GB.

  • Multi-GPU mode[restart]: If you have multiple graphics cards with identical video memory, you can enable this option and set the above option to 2 or the number of graphics cards.

【Video Output】 ​

  • Video output quality control (CRF): Constant Rate Factor (CRF) for video quality. 0=lossless (huge file), 51=low quality (small file). Default: 23 (balanced).

  • Output video compression preset: Controls the encoding speed vs. quality balance (e.g., ultrafast, medium, slow). Faster means larger files.

  • Custom FFmpeg command arguments: Custom FFmpeg command arguments, added before the output file argument.

  • Force software video encoding?: Force software encoding (slower but more compatible). Hardware encoding is preferred by default.

  • H.264/H.265 encoding: Video codec: libx264 (better compatibility) or libx265 (higher compression).

【Translation】 ​

  • Batch size (lines) for traditional translation: Number of subtitle lines per request for traditional translation.

  • Batch size (lines) for AI translation: Number of subtitle lines per request for AI translation.

  • Pause (s) after each translation request: Delay (in seconds) between translation requests to prevent rate-limiting.

  • Send full SRT format for AI translation: Send full SRT format content when using AI translation.

  • AI temperature for translation subtitles: AI models temperature,default is 0.2

  • AI translation with full original subtitles: The inclusion of complete original subtitles as AI context information will result in higher translation quality. [Important Note] 1. An advanced model supporting extremely long contexts must be used.

  1. Token consumption will increase several times.

【Dubbing】 ​

  • Concurrent dubbing threads: Number of concurrent threads for dubbing.

  • Pause (s) after each dubbing request: Delay (in seconds) between dubbing requests to prevent rate-limiting.

  • Save dubbed audio for each subtitle line: Save the dubbed audio for each individual subtitle line.

  • Azure TTS batch size (lines): Number of lines per batch request for Azure TTS.

  • The higher concurrent of EdgeTTS: The higher the concurrent voice-over capacity of the EdgeTTS channel, the faster the speed, but rate throttling may fail.

  • Retries after EdgeTTS failure: Number of retries after EdgeTTS channel failure

  • ChatTTS voice timbre value: ChatTTS voice timbre value.

【Alignment】 ​

  • Maximum audio speed-up rate: Maximum audio speed-up rate. Default: 100.

  • Maximum video slow-down rate: Maximum video slow-down rate. Default: 10 (cannot exceed 10).

  • Number of characters per line for CJK: Number of characters per line for Chinese, Japanese, and Korean subtitles; more than this will result in a line break

  • Number of words per line for Other: Number of words per line for subtitles in other languages; more than this will result in a line break

【ASR Settings】 ​

  • VAD: Speech probability threshold: VAD: Minimum probability for an audio chunk to be considered speech. Default: 0.5.

  • VAD: Max speech duration(s): VAD: Maximum duration (s) of a single speech segment before splitting. Default: 8s.

  • VAD: Min silence duration for split(ms): VAD: Minimum silence duration (ms) to mark the end of a segment. Default: 500ms.

  • Merge sub into adjacent sub if less than: If a subtitle's duration is less than this value in milliseconds, attempt to merge it into an adjacent subtitle. Default is 1000ms

  • no speech threshold: no speech threshold

  • temperature: temperature

  • repetition penalty: Increasing this value helps reduce repetitions

  • compression ratio threshold: Decrease this value helps reduce repetitions

  • hotwords: hotwords

  • Short sub will be merged if selected: Short subtitles will only be merged if this option is selected

  • Batch size for Faster-Whisper model recognition: Batch size for Faster-Whisper model recognition, larger values are faster but require more GPU memory, too large a size may cause GPU memory to be exhausted

  • Select VAD: Select VAD

  • Model for speaker separation: The model used for speaker separation. The default is the built-in model, supporting both Chinese and English. Pyannote is optional. If selected, you must have a token from https://huggingface.co and agree to the Pyannote licensing agreement. For details, please visit the URL for a tutorial: https://pvt9.com/shuohuaren

  • Your token from huggingface.co: Enter your token from huggingface.co. Otherwise, you cannot use Pyannote speaker separation. For details, please see the tutorial: https://pvt9.com/shuohuaren

  • faster/whisper models: Comma-separated list of model names for faster-whisper and OpenAI modes.

  • whisper.cpp models: Comma-separated list of model names for whisper.cpp mode.

  • CUDA compute type: CUDA compute type for faster-whisper (e.g., int8, float16, float32, default).

  • Recognition accuracy (beam_size): Beam size for transcription (1-5). Higher is more accurate but uses more VRAM.

  • Recognition accuracy (best_of): Best-of for transcription (1-5). Higher is more accurate but uses more VRAM.

  • Enable context awareness: Condition on previous text for better context (uses more GPU, may cause repetition).

  • Threads nums for separation: The more threads used for separation of human and background voices, the faster the process, but the more resources it consumes.

  • Convert Traditional to Simplified Chinese subtitles: Force conversion of recognized Traditional Chinese to Simplified Chinese.

【Whisper Prompt】 ​

  • initial prompt for Simplified Chinese: Initial prompt for the Whisper model for Simplified Chinese speech.

  • initial prompt for Traditional Chinese: Initial prompt for the Whisper model for Traditional Chinese speech.

  • initial prompt for English: Initial prompt for the Whisper model for English speech.

  • initial prompt for French: Initial prompt for the Whisper model for French speech.

  • initial prompt for German: Initial prompt for the Whisper model for German speech.

  • initial prompt for Japanese: Initial prompt for the Whisper model for Japanese speech.

  • initial prompt for Korean: Initial prompt for the Whisper model for Korean speech.

  • initial prompt for Russian: Initial prompt for the Whisper model for Russian speech.

  • initial prompt for Spanish: Initial prompt for the Whisper model for Spanish speech.

  • initial prompt for Thai: Initial prompt for the Whisper model for Thai speech.

  • initial prompt for Italian: Initial prompt for the Whisper model for Italian speech.

  • initial prompt for Portuguese: Initial prompt for the Whisper model for Portuguese speech.

  • initial prompt for Vietnamese: Initial prompt for the Whisper model for Vietnamese speech.

  • initial prompt for Arabic: Initial prompt for the Whisper model for Arabic speech.

  • initial prompt for Turkish: Initial prompt for the Whisper model for Turkish speech.

  • initial prompt for Hindi: Initial prompt for the Whisper model for Hindi speech.

  • initial prompt for Hungarian: Initial prompt for the Whisper model for Hungarian speech.

  • initial prompt for Ukrainian: Initial prompt for the Whisper model for Ukrainian speech.

  • initial prompt for Indonesian: Initial prompt for the Whisper model for Indonesian speech.

  • initial prompt for Malay: Initial prompt for the Whisper model for Malaysian speech.

  • initial prompt for Kazakh: Initial prompt for the Whisper model for Kazakh speech.

  • initial prompt for Czech: Initial prompt for the Whisper model for Czech speech.

  • initial prompt for Polish: Initial prompt for the Whisper model for Polish speech.

  • initial prompt for Dutch: Initial prompt for the Whisper model for Dutch speech.

  • initial prompt for Swedish: Initial prompt for the Whisper model for Swedish speech.

  • initial prompt for Hebrew: Initial prompt for the Whisper model for Hebrew speech.

  • initial prompt for Bengali: Initial prompt for the Whisper model for Bengali speech.

  • initial prompt for Persian: Initial prompt for the Whisper model for Persian speech.

  • initial prompt for Urdu: Initial prompt for the Whisper model for Urdu speech.

  • initial prompt for Cantonese: Initial prompt for the Whisper model for Cantonese speech.

  • initial prompt for Filipino: Initial prompt for the Whisper model for Filipino speech.