Real-time transcription and live captions

Real-time or batch: which mode you need

Real-time transcription puts words on screen while people are speaking, batch transcription produces the corrected record afterward, and most teams end up using both.

Real-time (streaming) transcription

Real-time transcription converts speech to text as it happens, typically within a second or two. That speed is the whole point: captions on screen while the speaker is mid-sentence. The trade-off is that no live system, human or AI, gets a second pass. Whatever the engine hears is what your audience sees, including misheard names, crosstalk and background noise. Live output is an aid in the moment, not a record.

Batch transcription

Batch transcription processes a complete recording after the fact. Sonix transcribes a one-hour file in about 5 to 6 minutes, and because the engine works with the full audio and its complete context, accuracy is higher, speakers are labeled and you can review and correct every line in the browser editor before anyone relies on it. This is the version that you publish, search, subtitle and cite.

Fast batch output is real-time enough for most teams, and it beats raw live text any time the transcript will be read, quoted or archived.

Speed means two things: latency and turnaround

Speed claims in transcription hide two different clocks, so here is what Sonix delivers on each.

Latency in real-time mode

Latency is the gap between a spoken word and its text on screen, and Sonix keeps it under 2 seconds. That is fast enough for captions to track a natural speaking pace in meetings, webinars and live events without the audience feeling the lag.

Turnaround in batch mode

Sonix processes recordings up to 10x faster than real time. A one-hour recording is typically done in about 5 to 6 minutes, and a 30-minute town hall takes about 3. Compare that with doing it yourself: the average person types 40 words per minute, so transcribing one hour of audio by hand takes 4 to 6 hours.

In practice the gap between live and fast batch is smaller than it sounds. End a one-hour recording, upload it, and the finished transcript exists before your follow-up coffee does. For any use where the words will be checked, quoted or kept, those few minutes buy you higher accuracy, speaker labels and an editable record.

Live captions or a fast transcript: how to choose

The test is simple: live captioning is for text people read during the session, and a fast transcript is for text they read afterward.

When you need live captioning

You need live captioning when the words must appear while people speak: accessibility for deaf and hard-of-hearing audiences, multilingual events, regulated broadcasts, or any session where waiting even five minutes defeats the purpose. That is what Sonix real-time transcription is for, and why it is rolling out first to enterprise teams running high-stakes live sessions. For a comparison of the tools available today, see our guide to the best live captioning software, or start with what live captioning is and how it works.

When you need a fast transcript

You need a fast transcript when the text is for afterward: subtitles, meeting notes, quotes, SEO and archives. Here batch wins on every axis except immediacy. Upload the recording as the session ends and Sonix returns a transcript you can correct, then export as SRT or VTT with Sonix automated subtitles and captions. Publishing a video with uncorrected live captions burned in is a shortcut audiences notice.

Most organizations end up pairing them: real-time for the audience in the moment, batch for everything the moment leaves behind. Standard transcription starts at $10 per hour of audio, and our guide to how much transcription costs has the full breakdown.

Streaming and batch run through one API

The Sonix API supports both modes, so one integration covers live sessions and recordings.

The real-time streaming API uses WebSockets for low-latency speech to text inside your own application, and it is available through the early access program. The batch side of the Sonix API is open now.

Batch is also how teams build near-real-time pipelines without streaming at all: push each recording the moment a session ends, poll for completion, and post the finished transcript to your knowledge base, CMS or compliance archive minutes later. Recurring meetings, webinar libraries and podcast backlogs run hands-free this way.