Live Transcription
How real-time transcription works in Kalima, from pressing record to a finished transcript.
Live transcription is the heart of Kalima. You press record, speak (or capture audio from your computer), and watch the words appear on screen within about a second, already separated by speaker, labeled by language, and ready to edit, share, or export. This page explains how it works from the moment you hit record to the moment you stop.
How live transcription works
When you record, Kalima sends your audio to its transcription engine continuously and streams the results straight back into the Studio. You don't wait for a file to finish processing, the transcript builds itself in real time as the conversation unfolds.
Preliminary text vs. confirmed text
You'll notice each phrase appears almost instantly and then settles a moment later. That's intentional and worth understanding:
- Preliminary (non-final) text shows up first, a fast, tentative result that may still change as you keep speaking.
- Confirmed (final) text replaces it once the engine has finished the phrase and is sure of the wording.
Because the engine waits briefly before finalizing a phrase, confirmed text can trail the live audio by up to about two seconds. This small delay is what keeps your transcript clean and readable instead of choppy and constantly re-shuffling.
Faint or lower-opacity words are the engine's way of flagging lower confidence. Treat them as words worth a second glance when you proofread, they're a hint, not a verdict.
Starting and stopping a recording
Open the Studio and run through quick setup, choose your audio source and microphone, add any expected languages, and turn on translation if you need it.
Press record. Kalima connects to the transcription engine and then begins capturing. Audio is captured from the instant you press record, so the opening words aren't lost while the connection completes.
Watch the transcript appear live. Words stream in, speakers are labeled, and any translation shows alongside the original.
Press stop to finalize the recording. Your audio is saved and your transcript is ready to play back, edit, and export.
A live latency indicator sits in the Studio while you record so you always know the session is healthy, more on that below.
For the most accurate transcript, give the engine a head start: pick a good microphone, choose the right audio quality mode, and add expected-language hints and session context when you know what's going to be discussed.
One session, many recordings
A session isn't limited to a single take. You can stop and start recording as many times as you like within one session, and each take is added to the same transcript on a continuous timeline. You can also pause and resume to take a short break without ending the current take.
This means meetings with breaks, multi-part interviews, or sessions where you step away all stay in one tidy transcript rather than scattering across separate sessions. Speaker numbering even stays consistent across every take in the session, so Speaker 2 in your third recording is the same Speaker 2 from the first.
You can run one recording at a time per account. Finish or stop your current recording before starting another.
The live latency indicator
While you record, a color-coded latency badge shows how quickly transcription results are arriving:
| Color | What it means |
|---|---|
| Green | Results are arriving fast, your connection and device are healthy. |
| Yellow | Moderate delay, usually a slower or busier network. |
| Red | Slow results, or the connection is interrupted or stale. |
Click the badge to expand it for average, minimum, and maximum timing plus connection details. Latency reflects your network and device, so a yellow or red badge usually points to a slow or unstable connection, or to a high-latency input device. Bluetooth headsets and virtual audio cables add their own delay, and Kalima will show a non-blocking device latency advisory when it spots one. Switching to a wired connection or a wired microphone is the quickest fix.
The latency badge is a measurement, not a control, it tells you how things are going so you can react, but it can't speed up your own network.
Automatic stop when your balance runs out
Recording time is metered against your balance, and Kalima makes sure you're never caught off guard:
- When you get low (under about two minutes of recording time remaining), a warning appears showing how much time is left.
- When your balance reaches zero, a short grace countdown appears and then recording stops automatically.
You can add more time or credits at any point to keep going, and you can always press stop yourself. If you record within a team or organization, available shared balances are drawn from automatically.
There are no surprise overages, you're warned before time runs out and the session stops cleanly when it's gone. To keep recording uninterrupted, top up before you hit the low-balance warning. See Minutes, Credits & Balance.
Built to survive interruptions
Live transcription needs a network connection, but a shaky one won't cost you your words. If your connection drops or transcription stalls, recording keeps going locally and Kalima automatically re-transcribes the missed stretch and merges it back into the transcript in the right place once you're reconnected. Recovered sections are clearly marked, and your audio is saved continuously as you go so a crash or tab close mid-recording doesn't lose it.
A brief Wi-Fi blip or a server hiccup is handled for you, keep recording and the transcript fills itself back in. For the full story, see Reliability & Recovery.
Plan limits
Live transcription is available on every plan. What varies is how much you can record per week and how long a single live session can run:
| Plan | Weekly transcription minutes | Max live session length |
|---|---|---|
| Free | 120 min | 90 min |
| Plus | 1,200 min | 180 min |
| Pro | 3,000 min | 300 min |
| Team | 3,000 min | 300 min |
| Enterprise | Unlimited | Unlimited |
See Plans & Pricing for the full comparison, and Usage & Quotas to track what you've used this week.
Frequently asked questions
Live transcription shows a fast preliminary result first, then refines it into the confirmed version a moment later. The confirmed text can trail the audio by up to about two seconds, which is normal, it helps avoid choppy, over-segmented results and keeps the transcript readable.
No. The spoken language is detected automatically, and the transcript handles mid-conversation language switches with per-segment language badges. For better accuracy, you can add expected-language hints if you already know what will be spoken. See Languages & Detection.
Yes. Each take is added to the same session and the timeline continues, so meetings with breaks or multi-part recordings stay in one transcript. Pause and resume keep the same take going without ending it.
Keep recording, your audio is saved locally and reconnection is automatic. Anything missed during the outage is re-transcribed afterward and merged into the transcript in the correct place. You'll see a reconnecting or restoring status while it works. More in Reliability & Recovery.
The badge reflects how fast results are arriving over your current connection. Yellow or red usually means a slow or unstable network, or a high-latency device. Try a wired network connection and a wired or built-in microphone, Bluetooth headsets and virtual audio cables add delay.
Yes. You'll get a low-balance warning when you're under about two minutes of remaining time, then a short grace countdown, after which recording stops automatically. Add more time or credits to continue. See Billing & Balance Issues.