The Scribe API transcribes audio and video in Fast, Batch, and Live modes. It can also return polished text when you enable Text Polishing.
Key features
- Processes common audio and video formats in Fast and Batch modes
- Streams real-time audio and transcription events in Live mode
- Provides Captions (partial transcription results) before a Live transcript item completes (Partial results feature in preview)
- Returns display text and transcript segments
- Includes optional segments, word timestamps, speaker labels, channel numbers, and detected languages, depending on the mode and configuration
- Supports mono and stereo inputs with optional channel separation
- Recognizes ten languages with the
zoom-scribe-multilingualmodel and 26 additional locales with thewhispermodel - Improves recognition of names and domain terms with custom vocabulary
- Produces optional Text Polishing output for
en-US - Preserves timestamps and source metadata from Zoom RTMS
Partial results feature in preview
Preview Period: Zoom may, in its sole discretion, temporarily allow usage of the partial results feature that is not counted, in whole or part, against Customer's available Zoom Build Platform credit allotment. At the conclusion of any such period, standard credit-usage calculation and credit consumption shall commence.
Processing modes
| Mode | Transport | Use case |
|---|---|---|
| Live | WebSocket | Real-time captions, agents, RTMS applications, and other streaming experiences. |
| Fast | Synchronous HTTP | Individual files that need an inline result. |
| Batch | Asynchronous HTTP | Long recordings and collections of files in object storage. |
All three modes support transcription, model selection, language configuration, and vocabulary customization. Text Polishing is available in each mode, with mode-specific response behavior.
For higher English accuracy, select the Pro option, Scribe API Pro High-Accuracy, with the model ID zoom-scribe-en-pro.
Speech configuration
Scribe does not create conversation summaries, meeting notes, or action items. Pass a completed transcript to the Summarizer API when your application needs those artifacts.
For recognition settings, language coverage, and model selection, see Speech configuration.
Speaker diarization and channel separation
Speaker diarization and channel separation describe different aspects of the audio:
diarizationidentifies speakers from characteristics of their speech and adds a speaker label to transcript segments.channel_separationprocesses the left and right channels of a stereo input independently and adds a channel number to each segment.
You can enable either or both options. For best diarization results, use recordings with fewer than 20 speakers.
Supported media formats
Fast and Batch modes accept these containers:
- WAV
- M4A
- MP3
- MP4
- WebM
Audio can be mono or stereo with a sample rate of at least 8 kHz. We recommend a sample rate of 16 kHz or greater.
Resources
AI Services are part of the Zoom Build platform. Use Zoom.ai to create a project, get credentials, and manage your AI Services plan.