Capture · Transcription
Transcribe calls on your computer, with no bot and no upload
Barside transcribes calls on your computer. Choose Transcribe a call, press Start recording, and your microphone and the sound your computer plays turn into a timestamped transcript in a note, labeled You and Others. Speech runs locally with Whisper, so no bot joins the call and no AI key is needed.
If you want to transcribe calls on your computer without sending the audio to an online service, Barside does the whole job on the machine in front of you. Choose Transcribe a call, press Start recording, and Barside records your microphone and the sound your computer plays. Every 30 seconds it turns the new audio into text with Whisper, a speech model that runs locally, and adds the lines to a note with a timestamp and a label: You or Others.
Nothing joins the call, nothing is uploaded, and you need no AI key and no account. Barside is free for Windows, Linux and Mac. This guide covers setup, clean transcripts and what to do after the call.
How call transcription works in Barside
Barside records a call as two-channel audio: your microphone on one side, and your computer's sound on the other. Your computer's sound is everything it plays, so the people on a Zoom, Microsoft Teams or Google Meet call, a call in your browser, a phone call you answer on your computer, or any other call app all arrive on that side. Because Barside listens to your computer instead of joining the meeting, it works with whatever app you call from, and nobody sees an extra participant.
The recording is handled in 30-second pieces:
- Pieces with no sound at all are skipped, so a long pause does not fill your note.
- Each piece with sound is first saved in the note as a small audio clip called Call audio, so the audio survives even if transcription fails.
- Whisper turns the piece into text on your computer while the call goes on. The assistant shows Transcribing locally… as it works.
- Barside compares how loud each side was and labels every part You or Others, with its time from the start of the call, like
[03:12] You: Let's lock the launch date for Friday.
The note is called Call transcript and goes into the workspace you are working in. Barside marks it as a transcript, so AI search and meeting follow-through know what it is. Rename it, move it or write around the lines, like any other note.
How to transcribe calls on your computer, step by step
- Install speech once. Open the assistant with
Ctrl+Shift+A, select Assistant settings (the gear at the top), and choose Install speech. It downloads Whisper and its English model, about 150 MB, and needs no AI key. - Choose a language, if you need one. English is ready to use. For other languages, choose Install more languages · 141 MB under Speech language, pick a language or Detect automatically, then choose Save assistant settings.
- Open the call setup. Choose Transcribe a call at the bottom of the assistant, click the microphone on the desktop bar, or press
Alt+Shift+Spaceand typecall. A card shows the note title and the labels: You (microphone) and Others (computer audio). - Tell the people on the call. The card reminds you: "Let call participants know before recording."
- Press Start recording. A banner says Recording call · mic + computer, and new lines appear in the note every 30 seconds.
- Stop when the call ends. Choose Stop & save in the banner, or Recording · Stop and save in the tray menu. Barside transcribes the last piece, and the note is done.
With an AI connected, you can also ask for it in words. Type or say "Transcribe this call into a new note in my Clients workspace", and the AI assistant prepares the recording in that workspace. Recording still starts only when you press Start recording.
What the transcript looks like
Each 30-second piece adds a line with the time it was recorded, its Call audio clip, and the labeled text. A short stretch of a client call might read:
[00:04] You: Thanks for making time. Can we start with the launch date?[00:09] Others: Friday works for us, as long as the pricing page is ready.[00:15] You: Then let's lock Friday. I'll send the recap by Thursday.
Times count from the start of the call, even when silent stretches were skipped, so you can find any moment in the audio. The speaker labels are bold in the note. Each Call audio clip plays in the note, with Trim and Share link under its player.
The text is ordinary note text. Fix a name, add a heading, or turn a promise into a checklist item and give it a due date, and it shows on your task board.
Speakers, languages and accuracy
You and Others
Barside labels by channel, not by voice. Your microphone is You, and everything your computer plays is Others. It does not tell the other people apart, so on a group call everyone but you is Others.
Without headphones, your microphone also picks up the other people from your speakers. Barside only calls a passage yours when the microphone is clearly louder, but a quiet reply right at a change of speaker can still get the other label. A headset gives the cleanest result.
35 languages, or automatic detection
English uses Whisper's English model, which comes with the speech download. The optional multilingual model (141 MB, downloaded once) adds automatic detection and 34 more languages: Arabic, Catalan, Chinese, Croatian, Czech, Danish, Dutch, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Malay, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Slovak, Spanish, Swedish, Tagalog, Thai, Turkish, Ukrainian, Urdu and Vietnamese.
If you choose a language before that model is installed, Barside transcribes in English and tells you so, with a pointer to Assistant settings.
Tips for a clean transcript
- Wear headphones, so your microphone hears only you.
- Silence other sounds. Computer audio includes everything your computer plays, such as music, videos and notification sounds.
- Check names and numbers afterwards. Barside uses Whisper's compact base models (
base.enfor English), which can mishear names, numbers and crosstalk. The audio clip next to each passage makes checking quick.
Who uses call transcription
- Consultants and freelancers keep a transcript of every client call in that client's workspace, next to the project's notes and bookmarks.
- Product and research teams record user interviews and quote people word for word later, with the time of each quote.
- Sales and account managers keep a record of what was promised on a call, and turn it into dated tasks.
- Remote teams transcribe stand-ups and planning calls in their own language.
- Anyone who types notes while talking can focus on the conversation instead and skim the transcript later.
After the call: summaries, tasks and search
Barside has no AI of its own, so these steps use the AI you connect in Settings → AI: Claude, ChatGPT, Gemini or OpenRouter.
- Meeting follow-through. Choose the Meeting follow-through button next to the note's title. It writes a summary, quotes the decisions, proposes tasks with the owners and deadlines people said, and drafts a recap you can edit. You check owners and reminder times before anything saves, and Barside never sends the recap for you. See meeting follow-up and action items.
- The meeting agent. In Agents & approvals, on the Meeting tab, tick After each call transcript, prepare a follow-up. When a transcription finishes, it prepares tasks, reminders and a recap email, and each one waits in your approval inbox. It is off by default. See AI agents that ask before they act.
- Ask your library. Weeks later, ask "What did we decide about the pricing page?", and Library answers come from your transcripts, with the passage cited as a source.
- Your daily briefing. Commitments you save from a call show up in the daily briefing until you finish them.
- Your AI app. Through AI connections, Claude Desktop, ChatGPT and other MCP apps can read the transcripts in the workspaces you share, and the ready-made Meeting follow-up prompt turns one into decisions and tasks.
None of this is required. Without a connected AI you still get the full text, the audio, and the search bar, which finds a transcript by any word in it.
Details that matter
- One at a time. The assistant records one call or one voice command at a time.
- Safe to close. Closing the assistant stops and finishes its recording, and quitting Barside stops and saves every recording first. Barside does not install an app update while a recording or transcription is running.
- Windows, Linux and Mac. On Windows 10 and 11, Barside captures your computer's sound through the system's audio loopback. On Linux it uses PulseAudio or PipeWire loopback. In Barside for Mac, Barside's own sound helper records the computer's sound on macOS 14.2 or later, and macOS asks once to allow Barside under Screen & System Audio Recording. Whisper runs there as one universal build for Apple silicon and Intel Macs, using Metal.
- Live audio only. Barside transcribes a call as it records it. It does not transcribe an audio file you already have. For videos in your notes, the Clips editor writes down every word on your computer too.
- Barside's own meetings. In Barside video meetings, live captions are written on each person's own computer when a host turns them on, and Barside can save a meeting note with the transcript and the chat when the meeting ends, so you do not need Transcribe a call there.
- Voice commands share the engine. Voice commands and the optional wake word use the same local Whisper, so one speech install serves Hey Barside voice commands too. On Windows, the wake word also needs the English (US) Windows speech pack.
Privacy: what stays on your computer
The audio, the text and the note stay on your computer. The speech engine is downloaded once from GitHub and its model from Hugging Face. After that, speech turns into text with no network at all. Call audio and transcripts are part of your library, so they are included in your backups. With Barside Plus, cloud backup encrypts them on your computer before upload, and cloud sync, if you turn it on, copies them to your other computers through Barside's server.
Otherwise, a transcript leaves your computer only when you give it to an AI. Meeting follow-through and the meeting agent send the transcript to the AI company you connected, and the assistant and Library answers send the excerpts they need; that company's data policy applies. An AI app you connect through MCP can read a transcript only in the workspaces you share with it. A Call audio clip leaves your computer only if you make a share link for it. Read how recordings and transcripts work for the details.
Computer audio can include other apps, so check what is playing before you start. Always tell people on the call that you are recording: recording rules differ between countries and states, and some require everyone's consent.
How it compares
Barside makes and keeps the transcript on your computer, and any AI step uses an AI account you control. If you are choosing between Barside and a dedicated note taker, the Otter.ai alternative and Granola alternative guides compare them feature by feature. To see every option side by side, read the best transcription apps and the best AI meeting assistants.
Works well with
FAQ
Questions and answers
Is call transcription in Barside free?
Yes. Barside is free, and transcription needs no AI key and no subscription. Speech turns into text on your computer with Whisper, after a one-time download of about 150 MB (141 MB more for languages other than English). Only the optional AI steps, like meeting summaries, use an AI account you connect.
Does Barside join my Zoom or Teams meeting as a bot?
No. Nothing joins the call. Barside records your microphone and the sound your computer plays, so it works with whatever call app you use, and the other people see no extra participant. Tell them you are recording before you start.
Can Barside transcribe calls offline?
Yes, the transcription itself runs offline. After the one-time speech download, Whisper turns the audio into text on your computer with no internet connection. The call app still needs its own connection for the call.
Can Barside tell who is speaking?
It labels each part of the transcript You (your microphone) or Others (the sound your computer plays), with its time from the start of the call. It does not tell the other people apart by name. Headphones give the cleanest labels, because your microphone then hears only you.
Which languages can Barside transcribe?
English is built in. Install the optional 141 MB multilingual model in Assistant settings → Speech language to transcribe 34 more languages, including Spanish, French, German, Portuguese, Chinese, Japanese, Hindi and Arabic, or to detect the language automatically.
Where are my call recordings and transcripts stored?
In the note, in Barside's library on your computer: the audio as Call audio clips and the text as labeled, timestamped lines. They are included in your backups. Barside does not upload them on its own: an AI feature sends the transcript, or the parts it needs, to the AI you connected, and a share link or Barside Plus cloud backup and sync upload only when you turn them on. See how recordings and transcripts work.
How do I get a summary and action items from a call?
Connect your own AI in Settings → AI, then choose Meeting follow-through next to the transcript's title. You get a summary, quoted decisions, proposed tasks with owners and a recap draft to edit, and you confirm owners and reminder times before anything saves. See meeting follow-up and action items.
Does call transcription work on Mac and Linux?
Yes. Barside runs on Windows 10 and 11, on Linux and on Mac. On Linux, the computer's sound comes from PulseAudio or PipeWire, and transcription works the same way. On a Mac, recording the computer's sound needs macOS 14.2 or later, and macOS asks once to allow Barside under Screen & System Audio Recording.