BATS - Convert2Speech

Also available in:

Deutsch

Let text be spoken in your voice – voice cloning via Nvidia GPU

Convert2Speech creates spoken-audio files from text – in a cloned voice. The engine is the AI model Chatterbox (Nvidia-GPU accelerated, 23 languages): you supply a short voice sample (WAV) of the desired voice, and BAT Server lets the desired text – from an SRT or VisTitle file, or even from existing spoken audio – be spoken in that voice. Ideal for audiobook passages, podcasts and voice-overs with a consistent timbre.

Adding the menu entry

“Convert2Speech” is added through the “Add function...” assistant (right-click in the BIN → Add function... → pick Convert2Speech - Create cloned Voice). The assistant’s questions:

  1. Language (“Please select the language.”): 23 languages – including Arabic, Danish, German, Greek, English, Spanish, Finnish, French, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Dutch, Norwegian, Polish, Portuguese, Russian, Swedish, Swahili, Turkish, Chinese.
  2. Voice sample (file dialog, WAV): the sample is the reference for the cloned voice. Recommended quality: at least 10 seconds, 24 kHz or higher, single speaker, no background noise, ideally professionally recorded – and the spoken content and emotion should match the desired result (e.g. audiobook tone for audiobook output). A button in the dialog directly opens EDIUS’s Voice-Over setup for recording.
  3. Max. stretch (0–50 %, default 10 %): to fit the sentences into the original’s time slots, the speech rate may be stretched up to this percentage. Recommended: less than 15 %; 0 % disables stretching.
  4. Source type: register the menu entry for VisTitle sources (video) or for audio sources.
  5. Add result to the BIN? (Yes → registered under __BAT Server\Speech / No → file only).
  6. Placement (BIN or timeline) and name (suggestion: “TTS <language> <voice file> <stretch>%”), then [Preview '.bat0' file] / [Add menu entry].

How to use it

  1. Right-click the source file (SRT or VisTitle file, or audio/video containing speech) → Generate Speech submenu → your Convert2Speech entry.
  2. If the source has no text yet (plain audio/video), BAT Server transcribes it first with Whisper (model large-v2) – SRT and VisTitle files are read directly.
  3. Chatterbox then produces the WAV file in the cloned voice – the time stamps follow the clip’s frame rate, so the spoken passages match the original timing. With “add to BIN”, the result lands under __BAT Server\Speech right away.

Good to know

  • Sample quality = result quality: noise, background music or changing speakers in the WAV sample are mirrored by the clone. A clean 10–30 s sample in the target language and target emotion is the biggest lever.
  • Stretch in moderation: sentences that “do not fit” into the time slot are stretched up to the set maximum – larger values audibly distort. Below 15 % the result stays natural sounding.
  • The sample’s path is baked in: the full path of the WAV file is in the .bat0. If the sample is moved, adjust the path in the .bat0 or recreate the entry.
  • GPU: Chatterbox runs on the Nvidia GPU – without a suitable GPU the entry is not really usable.

Managing the preset

Entries created via the assistant are user presets: the .bat0 file lives under C:\ProgramData\EdiusPowerTools\BAT-server\bat_files\BIN\ or …\bat_files\timeline\ depending on your choice, and can be freely edited, renamed or deleted (deleting removes the context-menu entry). Language, voice-sample path and stretch value are baked into the .bat0 – changes take effect from the next run.