Automatic transcription
Desktop Call Recorder turns your recorded calls into text you can read and search. Transcription runs on your own computer, like the recording itself. Your call audio is not sent to any transcription service.
How it works
- Record a call as usual. Nothing extra is needed.
- Transcribe it: press Transcribe on the recording, or let it happen automatically (see below).
- Select the recording to read the transcript next to the player.
Automatic or manual
After installation, transcription is manual. Press Transcribe next to any recording and it is transcribed straight away, with the progress shown. One transcription runs at a time. To have every new call transcribed for you, switch on "Transcribe new recordings automatically" in Settings, Transcription. It is off by default. It runs after each new recording finishes and never touches older recordings, which you can transcribe by hand whenever you need them.
If you delete a recording while it is being transcribed, the transcription stops and no transcript files are left behind. A recording still waiting its turn is skipped. If you rename a recording while it is being transcribed, the transcript is saved under the new name.
Choosing an AI model
How accurate the transcript is, and how much of your PC it uses, depends on the AI model. There are five, and you choose one in the Set up transcription window described below:
| Model | Download | RAM while transcribing | Notes |
|---|---|---|---|
| base (default) | ~142 MB | ~500 MB | Fastest, with basic accuracy |
| small | ~466 MB | ~1.2 GB | Clearly better, still fine on an ordinary processor |
| large-v3-turbo | ~834 MB | ~2 GB | Most accurate, slower on older PCs. Needs 64-bit Windows |
| parakeet-v3 | ~456 MB | ~1.5 GB | Fast and accurate for 25 European languages, including English, German, Russian and Ukrainian. Needs 64-bit Windows |
| sensevoice | ~153 MB | ~800 MB | Fast for Chinese, Cantonese and English. Needs 64-bit Windows |
All five run on the processor. You do not need a graphics card. The base, small and large-v3-turbo models know 99 languages. parakeet-v3 and sensevoice know fewer, as the table shows.
Languages of your calls
To get the right model for your calls, open Settings, Transcription and press Set up languages and model. The Set up transcription window opens with the languages of your calls, filled in from your Windows languages and keyboard layouts the first time. Add or remove languages until the list matches the languages you speak on calls. Both sides of every call are transcribed in these languages.
Below the languages are the models that cover all of them. The one that suits your languages and your PC best is first, marked Recommended. Press Use, or Download and use for a model that is not on your PC yet. Not now closes the window without saving anything.
With base, small or large-v3-turbo, a single call language is used as it is, and with several the app works out which one each person speaks. With none set, it detects the language for each side of the call from all the languages the model knows. parakeet-v3 and sensevoice find the language themselves.
Until you have done this, a card at the top of the main window offers the same button. Closing the card hides it for good. The button in Settings and the Improve link in the transcript panel stay, so you can change the languages or the model at any time.
Downloading a model
The base model downloads by itself in the background after installation. Any other model downloads when you choose it in the Set up transcription window and press Download and use. A notice at the top of the main window shows the download with its percentage, and you can close Settings while it runs. The model you had keeps transcribing until the new one is ready. Then the new one takes over, and Settings, Transcription shows it marked "downloaded". If a model will probably need more memory than your PC has, the app warns you, but the choice is still yours.
Your name in transcripts
Your side of the call is labelled "Me". To use your own name instead, type it in "Your name in transcripts" in Settings, Transcription. The new label is used for transcripts made from then on. Transcripts you already have keep the old one.
Re-transcribe and edit
The transcript panel has two buttons for a transcript you already have.
Re-transcribe runs the transcription again, for example after you download a better model or change your call languages. It replaces the old transcript and keeps the speaker names.
Edit lets you correct the text line by line, such as a misheard name or term. Press Save when you are done.
The other three buttons at the top of the panel find text in the transcript, copy the whole transcript, and rename the contact. They are explained on the search and rename page.
The line under the buttons shows your call languages and the model, for example "English · parakeet-v3". Before you set up your languages it starts with "Language auto-detected". The Improve link next to it opens the Set up transcription window.
Who said what
Every recording keeps the two sides of the call on separate stereo channels, you on the left and the other person on the right. Because of that the transcript can tell the two speakers apart and label every line.
Read along during playback
When you play a recording, the transcript follows the audio. The line being spoken is highlighted, and clicking any line jumps playback to that moment. More about this is on the player page.
Where transcripts are stored
Each transcript is saved next to its recording as two plain files with the same name: a text file you can read and a subtitle file (.srt). They are made on your PC and stay there. Your call audio is never sent to an outside transcription service.
The free version transcribes the first 3 minutes of each call. Pro has no limit.