Audio to text with speaker separation: how to transcribe a meeting recording on your computer

Nurbek Rakhmetov, founder of AitJaz · November 3, 2026 · 8 min read

To turn audio into text with speaker separation, you need a program that recognizes speech and tells the voices apart. In AitJaz, this is the Audio & video to text section: pick a recording, tick Separate speakers and start the recognition. The result can be a Word document where each line starts with Speaker 1, Speaker 2 or a real name. By default, the recording is processed on your computer, without an internet connection.

I’m Nurbek Rakhmetov, the founder of AitJaz. In version 0.3.8, released on October 12, 2026, we added speaker separation to transcripts. Before that, a meeting transcript had paragraphs and timestamps, but you had to remember yourself who suggested what. Now every line starts with a name. I tested the feature on my own recordings. Below, step by step, I show how to use it and what you get.

What speaker separation means

Speaker separation (in technical terms, diarization) means the program works out from the voice who is speaking at each moment, and labels every line. In an AitJaz transcript, the lines are labeled “Speaker 1”, “Speaker 2” and so on. In Word, each speaker gets their own color. A new speaker always starts a new paragraph.

After the transcript is ready, you can replace “Speaker 1” with a real name using the Speaker names… button, and the names are updated throughout the file. A fragment with timestamps looks roughly like this:

How to transcribe a meeting recording: step by step

1. Install AitJaz. Download the program for Windows 10 or 11 and run the installer. The first 7 days include all features, with no card and no sign-up.
2. Open Settings. Click the AitJaz icon next to the clock and choose Settings. The Audio & video to text section is the first one in the list.
3. Pick the recording. Choose the audio or video file of the meeting.
4. Set the language and the format. Choose the language of the speech, then choose the format to save: Word, text or subtitles. For meeting minutes, Word is the most convenient. If you want, tick Timestamps and paragraphs by pauses.
5. Tick Separate speakers. In the number-of-people field, leave Auto, or choose an exact number from 2 to 6 if you know how many people spoke.
6. Start the recognition. Progress is shown in percent along with the time left. You can stop at any point with Cancel. You can close the settings window: the work goes on, and dictation keeps working as usual.
7. Add the names. When the transcript is ready, click Speaker names… and replace “Speaker 1” and “Speaker 2” with the names of the participants. The file is saved again with the names.
8. Open the result. Open shows the file, and Show in folder shows where it is saved. If you don’t choose a folder, the file appears next to the recording.

💡 The first time you turn on Separate speakers, AitJaz downloads a voice model once, about 27 MB. This needs the internet. After that, the model works without it.

Which recordings you can transcribe

Almost any recording works: a meeting, a call, an interview, a lecture or a voice message.

• audio: mp3, m4a, aac, wav, ogg, opus, oga (this is how voice notes from WhatsApp and Telegram are saved), flac, wma, amr, aiff;
• video: mp4, mov, mkv, webm, avi, wmv, 3gp, m4v, mpeg, ts. The program takes only the sound from a video, so a recording of a video call works too.

The length of the recording is not limited. You can transcribe a five-minute stand-up or a meeting that lasts several hours. Files with Cyrillic letters in their names open too. All the transcription features are on the page Audio & video to text.

What the result looks like: Word, text or subtitles

How long transcription takes

On my computer, one minute of Russian speech is recognized in about 13 to 15 seconds. Speaker separation adds about 15 seconds for every 5 minutes of recording.

By these measurements, an hour-long meeting with speaker separation takes about 16 to 18 minutes. On another computer the time will be different, because it depends on the processor and on the language of the recording. The program itself will show the most accurate estimate when it starts working.

How to get a more accurate transcript

• Give the number of speakers. If you know that three people spoke, choose 3 instead of Auto.
• Put the microphone closer to the speakers. A laptop in the middle of the table records everyone better than a laptop in the corner of the room.
• Agree not to interrupt each other. Where people speak at the same time, check the text and the labels carefully.
• Turn on timestamps. They make it easy to find a disputed moment and listen to it again.
• Check names, sums and dates. Before you send the minutes, look at them separately.
• Tell the participants that the meeting is recorded. It is a matter of courtesy and of trust in the team.

Where the recording is processed and whether it goes to the internet

By default, AitJaz processes the recording directly on your computer. No internet connection is needed, and the recording is not sent anywhere. This matters for meetings, where people discuss contracts, money and plans that you would rather keep on your own computer.

Since version 0.3.8, there is also a second option, Cloud service, which works with your own API key. If you already have an API key from OpenAI, Google Gemini, Groq, OpenRouter, Together AI or Mistral, you can enter it in settings, and the recording will be processed by that service. In that case the sound of the recording goes to the cloud, and you pay the service yourself, according to its price list. The key is stored encrypted. If the service does not respond, AitJaz processes the rest of the recording on your computer. The section always shows where the recording will be processed.

Who needs transcripts with speaker labels

• Managers and secretaries for minutes of meetings and planning sessions: you see who suggested what and who is responsible for what.
• Journalists and researchers for interviews: questions and answers are already separated.
• HR specialists for recordings of job interviews.
• Teachers and students for lectures and seminars with questions from the audience.
• Video authors for subtitles of videos with several participants.

If you need to write text by voice rather than transcribe a finished recording, see the article How to type with your voice on a Windows PC.

Frequently asked questions