Anosic Speech Recorder

Dictate into VS Code. Record from your microphone, get a Whisper transcript, and have it on the clipboard ready to paste, all on your own machine.

Platform support

If you're on a Mac and want to help test, or want to know when it lands, open or follow an issue at github.com/anosic/speech-recorder.

Features

• Toggle Recording with Ctrl+Alt+R (Cmd+Alt+R on macOS), or by clicking the microphone in the status bar. The shortcut never moves keyboard focus, so you can dictate while focus stays in a chat input or an editor. To use a different key, click the keyboard icon in the panel title (or run Anosic Speech Recorder: Change Keyboard Shortcut…), which opens Keyboard Shortcuts filtered to Toggle Recording.
• Transcribe on stop. The whole recording is transcribed when you stop, and the model is loaded while you are still talking. Recordings that are effectively silent are reported as silence rather than transcribed, because Whisper tends to invent short phrases for silence.
• Copied automatically. With anosicSpeech.autoCopy on (the default) the transcript goes to the clipboard as soon as it is ready. The Copy button reads Copied! only while the clipboard holds exactly the full transcript.
• Cleared automatically. With anosicSpeech.autoClear on (the default) the transcript is cleared 10 minutes after each transcription (anosicSpeech.autoClearTimeoutMinutes), so dictated text isn't left behind. The clipboard is emptied too if it still holds the transcript. The status line shows when the clear will happen, and a new transcription restarts the timer.
• Sidebar panel (Anosic Speech Recorder, in the Explorer): Record / Stop, a level meter, the transcript, Copy, Clear (empties the transcript, and the clipboard too if it still holds the transcript; anything you copied since is left alone), and Transcribe WAV file… for existing recordings.
• Status bar item: a microphone when idle, REC m:ss while recording, a spinner while transcribing, and Copied while the clipboard holds the transcript. Hover for the full status; click to toggle recording.
• The keyboard icon in the panel title changes the shortcut, and the gear icon opens the settings. Logs are in Output › Anosic Speech Recorder.

Privacy and downloads

Audio is captured and transcribed locally; it is never uploaded. The speech model is downloaded once from Hugging Face on first use (the default, small, is about 249 MB) and cached in the extension's storage, or in anosicSpeech.modelCacheDir if you set it.

Transcripts are kept in memory only, never written to disk. By default they are cleared 10 minutes after each transcription, along with the clipboard if it still holds the transcript; the status line shows when. Clipboard history tools (such as Windows' Win+V history) may keep their own copy, which the extension can't reach.

What's in the package

Everything needed to record and transcribe ships in the extension package, except the speech model. The download is about 19 MB and unpacks to about 49 MB. Sizes below are for Linux x64; other platforms are similar.

The speech model is not in the package. It is downloaded on first use (see above) and is larger than everything in this table combined.

Settings

Requirements

A working microphone. The extension runs on the machine where your workspace lives (for remote and WSL windows, the remote side), and its package is platform-specific because it ships native audio and inference libraries.