Voice Transcriber

Linux dictation · v1.0 Flatpak · explicit provider boundary

Dictate the thought.
Keep the words.

Turn a spoken note, prompt, email, or draft into editable text on Linux. Voice Transcriber filters silence locally, shows exactly when completed speech segments go to Groq, and leaves storage under your control.

Install v1.0 Watch the workflow

No raw-audio recording or analytics. History is off by default. Inspect the app and its boundary before adding a key.

VOICE TRANSCRIBER

LIVE SESSION

■   Stop listening

Listening · speech detected (3.2s)

INPUT SIGNALUSB microphone

“The important part is getting the idea out before it disappears.”

Keep on top  ●Clear   Copy

Guided product tour · synthetic sample text

Know what happens before you press Start.

These reproducible previews mirror the current native GTK states. The application itself never inserts a simulated transcript.

Voice Transcriber ready state with microphone and privacy information visible
01 · Inspect. Open the desk and see the device and boundary before configuring a key.
Voice Transcriber active state with speech detected and input meter moving
02 · Speak. Local voice detection and the signal meter make capture state visible.
Voice Transcriber completed transcript ready for copy or export
03 · Use. Review the result, then copy, clear, or explicitly export it.

View the short animated tour · Reproduce these assets

The point

A small, dependable desk for when typing gets in the way.

There is no audio library to organise and no generic dashboard to navigate. Start a session, talk naturally, then put the words where they belong.

The short path

From microphone to useful text.

  1. 01

    Listen

    Choose a microphone and confirm its local signal before 30 ms frames enter a bounded real-time queue.

  2. 02

    Detect

    Voice activity detection runs on your device, so silence never becomes a transcription request.

  3. 03

    Transcribe

    Completed segments cross the visibly selected provider boundary through a small bounded request pool.

  4. 04

    Use it

    Review the growing transcript, then copy it or export a local UTF-8 text file.

An honest boundary

Local capture. Cloud transcription. No ambiguity.

Stays on your device

  • Raw microphone frames in a bounded memory queue
  • The active transcript until you clear, copy, or export it
  • Window preferences and local settings

Sent when you speak

  • Completed speech segments, encoded in memory as WAV
  • Your configured language or translation request
  • Nothing is sent while waiting in silence

Read the complete data flow and provider boundary

Release proof, not a trust badge

Checksum. Test report. SBOM. Provenance.

Every v1 release maps its source tag to one Flatpak and publishes the evidence needed to inspect it. Automated package proof is kept separate from the real-microphone compatibility reports still being collected.

Inspect the v1 release · Read the support boundary · Choose a contribution

Checksum-verified x86_64 Flatpak

Install v1.0, then inspect it before adding a key.

terminal

curl -LO https://github.com/OthmaneBlial/audio-capture/releases/download/v1.0.0/voice-transcriber-1.0.0-x86_64.flatpak
curl -LO https://github.com/OthmaneBlial/audio-capture/releases/download/v1.0.0/voice-transcriber-1.0.0-x86_64.flatpak.sha256
sha256sum --check voice-transcriber-1.0.0-x86_64.flatpak.sha256
flatpak install --user ./voice-transcriber-1.0.0-x86_64.flatpak
flatpak run io.github.othmaneblial.audio_capture

The primary package targets x86_64 Linux with GTK 3 and the system PipeWire/PulseAudio route. The UI, device picker, diagnostics, and privacy explanation work without a key; Start stays blocked until the selected provider is ready. Read the support matrix or use the source development path.

Built for the moment

Keep the conversation. Lose the friction.

Explore the source