What are we transcribing today?
Volocal transcribes speech with AI models on your own computer. Spots the languages. Marks the speakers. Fixes mistakes. Handles any audio format, video or an online stream. Free. Without losing control or privacy.
Version 1.2.25 for Windows 10 and 11
Recordings are never sent anywhere. Once it is set up, you can work offline.
Speech is split between the speakers, and you give them their names. The minutes then read on their own.
Corrections, translation and summaries all happen locally. The models always run on your own computer.
You always see what is playing. No scrubbing, no hunting. One click is enough.
Not a penny on transcripts this year.
You can search the text and correct whole sentences or single words.
Podcasts from YouTube go on the blob. blog.
Volocal marks the spots it was unsure of, so you know where to look before the transcript goes any further.
Voice notes go to Petr.
If a recording has a second language in it, Volocal notices and fills the transcript in. Ideal for recordings made with a live interpreter.
Record from a microphone, paste a link to an online service, or drop a file into the window.
Text to TXT, Markdown, SRT, VTT or JSON; audio to MP3, M4A or WAV.
Mark a passage in the transcript, right-click, and save only that: audio cut straight out of the recording, or its text and subtitles.
Last year we paid forty thousand for transcripts.
Pin them to the exact moment in the recording, and the panel keeps them together.
Volocal watches the folder you choose, transcribes what lands in it and files it away.
Transcripts go where they belong. The audio files on disk stay where they are.
From three-quarter speed to double, depending on how fast somebody talks.
Export SRT or VTT subtitles and use them with the original video or audio.
No telemetry. No crash reports. No account. Recordings, transcripts, notes and settings sit safely on your own disk. Volocal reaches the internet exactly three times:
It downloads the tools and models from their authors’ own sites.
Only the request goes out; the audio comes back.
It only ever asks. You start the download.
A safe, blazing-fast stack is the foundation. A proper GPU turns it into a rocket, but it will run on an older CPU too.
The core is in blazing-fast Rust, the interface in React over the system WebView, the archive in SQLite. Volocal is small and starts instantly.
Whisper runs on your own hardware, not on somebody else’s server behind a browser tab. Silero VAD sorts the speech from the silence.
You choose the first time you open it, and you can switch the model whenever you like. Transcripts already made stay as they are.
Volocal works out whether CUDA or Vulkan fits; Vulkan uses integrated graphics too. A GPU is not required — every model runs on the processor as well, only slower.
That is what the fast model needs on a graphics card. On a processor, about eight minutes. The accurate model takes four minutes on a card and thirty-five on a processor.
The models take 700 MB to 1.7 GB, depending on what you choose. Nothing beyond Windows 10 or 11 is needed.
Volocal is a powerful helper, but there are things it cannot do.
They are good enough for following the text, but in a long sentence with a pause in the middle they can drift a little. The timings are tighter with the bigger model.
Volocal tells individual speakers apart without trouble, but in a crowd of voices it picks the louder one. Where it gets messy, a manual fix helps.
Structured conversations with several people, yes. A chaotic panel discussion, not really. Two voices usually come out right; a crowd talking over each other is still a challenge.
Volocal takes one recording at a time. Add ten of them and they wait in a queue.
No. It is an ordinary file, with whatever protection your user account has.
Three steps lead from the download to a transcript.
It installs for you alone and needs no administrator rights. If you want the newest version at all times, switch automatic update checks on in the settings.
The installer is not signed. You will see Windows protected your PC. Choose More info → Run anyway.
That is 700 MB to 1.7 GB, depending on what you choose. Then you drop a recording into the window and Volocal does the rest.
Ready for more? Connect your people, data and knowledge with an AI-ready ecosystem of tools.
Version 1.2.25 for Windows 10 and 11. The author uses it every day.