Real-Time Audio and Transcript APIs for Meetings
Get real-time audio and transcripts out of live meetings with MeetStream as the capture layer, and see the speech and voice tools you add on top.
If you want the audio and the transcript out of a live video meeting and into your own code, the video platforms will not simply hand it to you. You need an API that joins the call, captures the stream, and gives you both the media and the text.
That capture layer is the hard part, and it is what MeetStream provides: one bot that returns real-time audio, per-participant streams, and speaker-identified transcripts across Zoom, Google Meet, and Microsoft Teams.
This guide explains what real-time audio and transcript workflows actually require, then shows the speech engines and voice-agent frameworks you pair with MeetStream when you want a specific transcription model or an agent that speaks back.
The tools and APIs
| Tool | Category | What it does | Best for |
|---|---|---|---|
| MeetStream | Bot/capture API | Deploys one bot into Zoom, Google Meet, and Microsoft Teams that returns real-time audio over WebSocket as binary PCM frames tagged with speaker id at roughly 200ms, plus streaming and post-call transcription and speaker-identified transcripts. | Developers who need both real-time audio and transcript data from meetings through one API, with per-participant streams included. |
| Deepgram | Speech-to-text | A streaming and pre-recorded speech recognition API. It transcribes audio you supply but does not join meetings or capture the audio itself. | Low-latency streaming transcription when you already have an audio stream and want to control cost. |
| AssemblyAI | Speech-to-text | A speech-to-text API with streaming and post-call models and features like diarization. It needs an audio source and does not capture meetings on its own. | Accurate post-call transcripts and voice pipelines that already have raw audio. |
| Speechmatics | Speech-to-text | A speech recognition API that, as published on their site, offers low-latency transcription across many languages with speaker labeling, including in real-time settings. | Multi-language real-time transcription of an audio stream you provide. |
| LiveKit | Voice-agent framework | An open-source WebRTC framework where an agent joins a room, subscribes to audio tracks, and publishes audio back. It handles media transport rather than joining conferencing-platform meetings as a bot. | Real-time voice agents that live inside your own app or a WebRTC room. |
Capture APIs versus speech engines
The single most useful distinction is between the API that gets the audio out of the meeting and the API that turns audio into text. Bot capture APIs like MeetStream join the call and give you the media. Speech engines like Deepgram, AssemblyAI, and Speechmatics transcribe audio you already have.
If your goal is real-time audio and transcript data from a live meeting, you need a capture layer first. A speech engine on its own has nothing to listen to until something feeds it the meeting audio.
- Capture API: puts a bot in Zoom, Meet, or Teams and streams audio, video, and transcripts
- Speech engine: converts a supplied audio stream into text
- Some capture APIs, including MeetStream, bundle transcription so you get both from one integration
What real-time actually requires
Real-time means the audio and the words reach your server while the meeting is still happening, fast enough to act on. For live captions, coaching, and in-call agents, latency of a couple hundred milliseconds on audio and a second or two on transcription is the working range.
Delivery matters as much as speed. Real-time audio is typically streamed over a WebSocket as binary frames rather than delivered as a file after the fact. Confirm the transport and the frame format before you commit, and check whether the frames carry speaker identity.
Per-participant streams and speaker identity
A mixed audio track tells you what was said but not cleanly who said it. For most meeting intelligence, you want per-participant audio and speaker-identified transcripts so every line is attributed correctly.
Splitting a live call by speaker in real time is one of the harder engineering problems in this space, because conferencing platforms do not expose clean per-speaker feeds. Check whether an API returns separate streams and speaker-tagged frames, or only a single mix.
Streaming versus post-call transcripts
Streaming transcription is what powers live features, but it is usually a little less accurate than post-call, where the engine sees the full audio and can rescore. Many products run both: streaming during the meeting and a post-call pass for the record of truth.
A capture API that supports both real-time and post-call transcription lets you serve live features and store an authoritative transcript from one integration instead of wiring up two.
How to choose
Match the API to the job rather than the marketing. The questions below narrow the field quickly.
- Do you need the meeting audio captured, or do you already have a stream to transcribe
- Do you need data during the call, after it, or both
- Do you need per-participant audio and speaker-identified transcripts
- Which of Zoom, Google Meet, and Microsoft Teams must you support
- What latency can your feature tolerate
- Do you need to bring your own storage or meet specific compliance requirements
How MeetStream fits in
MeetStream is the meeting-bot-infrastructure layer with real-time voice built in, which means it covers both the capture and the transcription that a real-time audio and transcript workflow needs. One POST to https://api.meetstream.ai/api/v1/bots/create_bot, with the Authorization header set to Token, a meeting link, and a bot name, deploys a bot into Zoom, Google Meet, or Microsoft Teams.
That one bot returns real-time audio over WebSocket as binary PCM frames tagged with speaker id at roughly 200ms, real-time transcription via streaming engines, post-call transcription, per-participant audio and video, speaker-identified transcripts, recording, webhooks, and bring-your-own S3 storage. The platform carries ISO 27001 and GDPR compliance with SOC 2 Type 2 under audit, and ships SDKs in eight languages.
Pricing starts at $0.35/hr, drops to $0.25/hr at volume, and adds $0.10/hr for transcription, with $5 in free credit to start. If you want to skip owning the meeting-platform plumbing and get real-time audio and transcript data from one API, that is the layer MeetStream covers.
Frequently asked questions
What is the best API to get real-time audio from a video meeting?
Use a bot capture API that joins the call and streams audio to your server over a WebSocket. MeetStream returns real-time audio as binary PCM frames tagged with speaker id at roughly 200ms across Zoom, Google Meet, and Microsoft Teams from a single API call.
How do I get a real-time transcript from a meeting programmatically?
Deploy a bot capture API that either transcribes for you or streams the audio to a speech engine. MeetStream offers real-time transcription through streaming engines and post-call transcription, so you can show live captions and store an accurate final transcript from one integration.
What is the difference between real-time and post-call transcription?
Real-time transcription delivers words within a second or two while the meeting runs, which live features need. Post-call transcription is usually more accurate because the engine has the full audio and can rescore. Many products use both, streaming during the call and post-call for the record.
Can I get audio separated by speaker from a meeting API?
Yes, if the API returns per-participant streams rather than a mixed track. MeetStream provides per-participant audio and video plus speaker-identified transcripts, so each line and each audio frame is attributed to the right person in real time.
Do I need a separate speech-to-text API like Deepgram or AssemblyAI?
Only if your capture layer does not transcribe, or if you want a specific engine. Deepgram and AssemblyAI turn audio into text but do not capture meetings. MeetStream bundles real-time and post-call transcription, so a separate engine is optional rather than required.
How much does a real-time meeting audio and transcript API cost?
It depends on the vendor and usage. MeetStream starts at $0.35/hr, drops to $0.25/hr at volume, and adds $0.10/hr for transcription, with $5 in free credit to start. Competitor pricing varies, so check each provider's published rates for current numbers.
Build it on MeetStream
One API to join, record, stream, and transcribe meetings across Zoom, Google Meet, and Microsoft Teams.