Transcription overview
Compare real-time and async transcription, explore providers and features, and start with quickstarts for live transcript events or fetching a completed transcript.
Use transcription to turn meeting audio into timestamped text for summaries, action items, search, coaching, compliance review, and live captions.
A transcript captures what was said and can include speaker labels and other metadata. Transcription is not enabled by default: configure real-time transcription before recording, or create a post-meeting transcript from a completed recording.
Quickstarts
Create a meeting bot with real-time transcription and receive transcript events during the meeting.
For bots and desktop SDK. Create a transcript from a completed recording, wait for processing to finish, and fetch the result.
Real-time transcription with local audio using the Desktop Recording SDK
Capturing audio locally instead of using a meeting bot? Follow the Desktop Recording SDK real-time transcription guide.
Transcription types
Choose when you need transcript data first. Then choose how to capture the meeting: with a meeting bot or the Desktop Recording SDK.
| Type | When to use it | How it works |
|---|---|---|
| Real-time transcription | You need live captions, transcript updates, alerts, or automations during the meeting. | Configure a transcription provider and a real-time endpoint before recording. Your application receives transcript events during the meeting. |
| Post-meeting (async) transcription | You only need the transcript after the meeting, for tasks such as summaries, action items, or search. | Create a transcription job for a completed recording, wait for completion, and retrieve the transcript. Works with recordings from meeting bots or the Desktop Recording SDK. |
Real-time transcription also makes the full transcript available after the meeting ends. Fetching that transcript does not require creating a separate post-meeting transcription job.
Implementation guides
| Capture method | Real-time transcription | Post-meeting transcription |
|---|---|---|
| Meeting bots | Receive live transcript events | Create and fetch an async transcript |
| Desktop Recording SDK | Receive live transcript events | Create and fetch an async transcript |
Speech-to-text provider options
A speech-to-text provider (the service that converts audio into text) generates your transcript. Recall.ai Transcription is the recommended option and requires no third-party provider setup. You can also use one of the third-party providers below.
For a third-party provider, add its API key in the Recall.ai dashboard by following its setup guide. Recall passes provider settings through directly, and the provider determines field defaults. Contact the provider's support team for guidance on accuracy or provider-specific configuration.
Use the provider guides to review configuration options for your chosen transcription type. Language support, defaults, and available features vary by provider.
| Speech-to-text provider | Setup | Post-meeting guide | Real-time guide |
|---|---|---|---|
| Recall.ai Transcription (recommended) | Not required | Post-meeting | Real-time |
| ElevenLabs | Setup | Post-meeting | Real-time |
| Deepgram | Setup | Post-meeting | Real-time |
| AssemblyAI | Setup | Post-meeting | Real-time |
| AWS Transcribe | Setup | Post-meeting | Real-time |
| Rev | Setup | Post-meeting | Real-time |
| Speechmatics | Setup | Post-meeting | Real-time |
| Soniox | Setup | Post-meeting | Real-time |
| Grok (xAI) | Setup | — | Real-time |
Key transcription features
Choose features based on your capture method, transcription type, and provider. Follow each feature guide for supported configurations and limitations.
| Feature | What it does | Learn more |
|---|---|---|
| Diarization | Identifies who said what. Speaker labels can map to known meeting participants or appear as generic labels, depending on the method and provider. | Diarization |
| Multilingual transcription | Supports meetings with one or more spoken languages. Language detection and code-switching (switching languages during speech) support vary by provider. | Multilingual transcription |
| Provider data | Exposes raw provider-specific fields that are not included in Recall.ai's normalized transcript format. Not available for Recall.ai Transcription or when diarization.use_separate_streams_when_available: true. | Access provider-specific fields |
| Partial transcripts | Delivers interim t |
Updated 1 day ago