Adobe Speech to Text for Premiere Pro

About this release

Transcription used to be the tax on any interview based edit: someone sat with the footage and typed. This component removes that step by running speech recognition against the audio on the timeline and producing a transcript that is linked to time, so clicking a word moves the playhead to the moment it was spoken and searching the transcript is the same as searching the footage.

The models run on the machine rather than being sent anywhere, which matters for embargoed or confidential material and also means transcription works with the network unplugged. Processing speed depends on the processor, but a long interview generally finishes in a fraction of its runtime, and a batch of clips can be queued to run while other work continues.

From the transcript, captions are one action away. The text converts into a caption track with sensible line breaks and timing, and from there the styling, positioning and safe area handling are ordinary caption work on the timeline. Captions burn into the picture or export as a sidecar file for platforms that expect one.

Speaker labelling separates voices in a multi person recording so an interview reads as a conversation rather than a wall of text, and the transcript itself exports as plain text or as a subtitle format for anyone who needs to work on the wording outside the editor. Corrections made in the transcript propagate into the captions generated from it.

What it does

Transcription on the machine

Recognition runs locally, so confidential footage never has to leave and no connection is needed.

Transcript linked to time

Clicking a word jumps the playhead, and searching the text searches the footage.

Captions in one step

The transcript converts into a styled caption track with sensible breaks and timing.

Speaker labelling

Separate voices are identified so multi person recordings read as a conversation.

Batch processing

Queue a folder of clips and let recognition run while the edit continues.

Text and subtitle export

Transcripts leave as plain text or as standard subtitle files for outside work.

Changed in this version

• Recognition accuracy improved on accented speech and noisy locations.
• Additional spoken languages added to the model pack.
• Faster processing on high core count machines.
• Speaker separation handles overlapping dialogue better.
• Caption line breaking reworked for more natural reading.

System requirements

Install order

1. Close the editing application before installing.
2. Unpack the archive to a short path.
3. Run the installer, which places the language models in the shared component folder.
4. Reopen the editor and confirm the languages appear in the transcription panel.

Worth knowing before you start

File details