How to transcribe a Hebrew recording, step by step

You have a recording — a team meeting, a client call, a lecture, an interview — and you need it as text. Typing it yourself costs roughly four to six hours per hour of audio. This is the short version of doing it with software instead: which file to upload, what to ask for, and what to check before you send the result to anyone.

Step 1: pick the right file

The single biggest lever on the result is the file you start from, and it costs nothing. Three rules:

• Use the original. Not a screen recording of the playback, not a re-compressed copy sent through three apps.
• Prefer audio-only when you have the choice. An 500MB-limit goes much further with an M4A than with a 4K screen capture, and the transcript is identical.
• Keep every channel. If your meeting tool offers a separate audio track per participant, that recording is the best input any transcription system can get.

Supported inputs are the ordinary ones: MP3, WAV, M4A, AAC, OGG/OPUS, FLAC, AMR, 3GP, WMA, AIFF, CAF, and video: MP4, MOV, WEBM, MKV, AVI. Anything up to 500MB.

Step 2: upload it

1. 1Open a new meeting. From the dashboard, choose to upload a file, or record straight from the browser if the conversation is happening now.
2. 2Drop the file in. Upload starts immediately; large video files take longer to send than to transcribe.
3. 3Say what you want from the summary. This is the step most people skip. "Focus on pricing objections and what we promised to send" produces a different, better summary than the default. It costs one sentence.
4. 4Close the tab if you like. Processing continues without you. The meeting appears in your list when the transcript is ready.

Step 3: read what comes back

You get three things, and they are useful in different ways:

The transcript is split by speaker with timestamps, so every line can be clicked back to the audio. The summary is written from the transcript text, and can be regenerated with a different instruction without transcribing again.

Step 4: name the speakers once

A fresh transcript labels voices as Speaker 1, Speaker 2, and so on — the system heard a change of voice, not a name. Rename them once and the names propagate through the transcript and the summary. Recognising the same person in a future meeting is a separate feature: it relies on a voiceprint, which is biometric data, so it is off by default and only applies to speakers whose consent was recorded. There is a longer explanation in speaker diarization, explained.

Step 5: export, or don't

You can export to PDF, Word or PowerPoint, or send a share link instead. A link is usually better: it stays current if the transcript is corrected, and a public preview link serves only the first 20% of the text — enforced on the server, not hidden with styling.

What to say in the room

One sentence at the start of the meeting handles almost every situation: "I'm recording this and running it through a transcription service, so we'll have the decisions in writing." In Israel you may record a conversation you take part in; recording one you are not part of, without consent, is a criminal offence. If someone objects afterwards, delete the meeting — deletion removes the stored files with it.

The honest limits

Distance from the microphone, room echo, and two people speaking at once degrade every speech system that exists. Names of people and companies, and long strings of digits, are where mistakes cluster even in good audio. That is why the transcript is editable, why speakers get named once, and why the timestamps exist. If a free trial disappoints you, listen to the first thirty seconds of your own recording before blaming the software — most of the time the answer is audible there.