BATS - Speech2Player ✂

Also available in:

Deutsch

Finding one specific line of dialogue in a long clip, zooming into the right moment and re-setting IN and OUT by hand can quickly turn into tedious scrubbing work in the player. Cut by Text (also known as Speech2Cut) turns the workflow around: the BAT Server transcribes the clip's audio track with Whisper AI down to word level, and then opens a transcript tab in the EDIUS Player. You simply select the text you are interested in – the player opens the clip at the corresponding position and sets the IN/OUT points automatically. Optionally, the passage is played back automatically (on loop) so you can verify it right away.

How it works

The workflow has three steps:

  1. Transcription in the background: the BAT Server extracts the audio track of the selected clip and lets the Whisper engine recognize every word with its exact time position. The result is an SRT file (one line per word) that is saved next to the source file with the suffix _stt-tabui. If that file already exists – e.g. from a previous run – the transcription is skipped and the result is re-used.
  2. Transcript tab in the player: once the job is finished, BATS automatically opens a tab in the EDIUS Player showing the full transcript. Every word is linked to its exact time position. Pauses in the speech are visualized: short pauses (over 0.5 s) as a dotted marker in the text, longer pauses (over 3 s) as a line break – so you can orient yourself in the text immediately.
  3. From text to marks: you select the passage you need in the transcript – a single word, a sentence, or a whole paragraph. BATS maps the selection to the corresponding time range, opens the clip in the EDIUS Player and sets IN and OUT automatically to the marked text segment. If the Loop Play option is enabled in the tab, the player plays the passage on loop automatically. You can then fine-tune the IN/OUT points by a few frames – or simply drag the clip into the timeline with the marks already set.

If you switch to a different tab in between and later come back to the transcript tab, BATS reopens the clip in the player at the marked position automatically.

How to use it

One-time: create the preset

  1. Right-click a clip in the BIN and choose "Add function…" – this opens the BAT Server add-in panel where the pre-installed extensions can be selected.
  2. Select the Cut by Text preset – there is one variant each for the Intel engine (runs on Intel NPU, Intel GPU, or CPU) and the Nvidia engine (Whisper-Faster on the Nvidia GPU).
  3. In the assistant dialogs, set the language (auto-detect or a specific language), the hardware for the calculation, the model size, where the menu entry should appear (BIN or timeline), and the preset name (default e.g. "Tab English, Model Medium, CPU").

Per clip: transcribe and mark

  1. Right-click the clip – the preset appears in the context menu under the Cut by Text submenu. In the standard setup the preset is registered for audio clips in the BIN.
  2. Start the preset. The BAT Server runs the job (progress is visible in the BAT Server panel) – if an SRT already exists, the tab opens directly without transcribing.
  3. In the transcript tab that opens, select the passage you want – the player sets IN/OUT on the clip accordingly, and with the Loop Play option enabled it plays the passage on loop.
  4. Fine-tune the IN/OUT points if needed, then drag the clip into the timeline.

Options

The engines

  • Nvidia (Whisper-Faster): runs on the Nvidia GPU and is particularly fast. It supports the models from Tiny to Large (V3) plus the very fast Turbo model. Optionally you can enable the high-precision float16 mode – it is required on RTX 5000 series GPUs and newer.
  • Intel (OpenVINO Whisper): runs on the Intel NPU, the Intel GPU, or the CPU (AMD included) depending on your system – so it works without a dedicated Nvidia card as well. It supports the models from Tiny to Large (V3).

Model sizes: from Tiny (very fast, coarse accuracy) through Base, Small and Medium to Large (V2/V3) (highest accuracy, longer processing time) – plus Turbo with the Nvidia engine. Medium is the recommended choice for most cases; for more difficult or quiet audio, Large is worth a try.

Language: either let the engine detect it automatically or fix one language – over 90 languages are available (German, English, French, Spanish, Japanese, Korean …). Since the language is fixed per preset, it is easy to create several presets – one per language.

Good to know

  • The SRT file is saved next to the source file with the suffix _stt-tabui and is re-used on every subsequent run of the preset – so re-runs finish instantly. If the clip's audio content has changed, delete the SRT file to force a fresh transcription.
  • Only one transcript tab per file can be open at a time: if a second tab is started for the same file, BATS shows a notice page ("Instance Already Running") and brings the existing tab to the front.
  • The "Open in Edge Browser" button opens the tab in a separate Edge window as well – handy when the embedded player view is too small for the transcript text.
  • Enable the Loop Play checkbox in the tab before you make your text selection – then the player plays the marked passage on loop; otherwise it plays it once.

Managing the presets

The presets you create show up in the clips' context menu under the "Cut by Text" submenu with the name you chose. You can rename, temporarily disable, or delete the entries, and create as many variants as you like – e.g. one per language or model – so your favorites are always directly available in the context menu.