Verba: The Free Voice Dictation Extension for VS Code

Voice control has long been part of everyday life. On the smartphone, in the car, in the smart home. But in the development environment? There, most people still type every word by hand. It doesn’t have to be that way.

With Verba, there is now a VS Code extension that brings speech directly into your editor. Record, transcribe, refine, insert. All in one step, without leaving VS Code.

Why Voice Dictation for Developers?

Developers don’t just write code. They write commit messages, documentation, code reviews, emails, Slack messages. All of this can be dictated faster than it can be typed. Especially with longer texts, voice dictation not only saves time but also spares your wrists and your concentration.

The problem: the existing tools don’t fit into the developer workflow.

The Market: Expensive and Off the Mark

Tools like Whisper Flow or SuperWhisper do offer solid speech recognition, but they are:

• Paid with monthly subscriptions
• System-wide tools that know nothing about the editor context
• Not integrated into VS Code – the text ends up somewhere, but not where you need it

For developers who spend their entire working day in Visual Studio Code, that is unsatisfactory. You need a tool that understands where your cursor is, whether you are working in the editor or the terminal, and what kind of text you need right now.

Verba: Native VS Code Integration

Verba is designed differently. It is not a system-wide dictation tool that happens to work in VS Code as well. It is a native VS Code extension, built for exactly one audience: developers in VS Code.

How It Works

1. Cmd+Shift+D (Mac) or Ctrl+Shift+D (Windows/Linux) – a Quick Pick menu appears
2. Choose a template (the first time; after that, the last template is reused automatically)
3. Speak
4. Same key combination – recording stops
5. The result appears at your cursor position

That’s it. No switching windows, no copy-paste, no cleaning up filler words.

What Happens Under the Hood

Verba uses a pipeline of two services:

• Deepgram Nova-3 transcribes the audio recording into text – fast and accurate
• Anthropic Claude cleans up and structures the text based on the chosen template

The template determines how Claude refines the transcribed text. For a commit message, the text becomes a concise, convention-compliant message. For free text, filler words are removed and sentences are smoothed out. For JavaDoc, structured documentation is produced.

The processing runs as streaming directly in VS Code – a live character counter in the status bar shows the progress. You can cancel at any time.

New Features at a Glance

Detect Self-Corrections Automatically (Course Correction)

Anyone who dictates sometimes corrects themselves mid-sentence: “Let’s meet tomorrow, no wait, on Friday at ten.” Verba detects such self-corrections automatically and outputs only the final text: “Let’s meet on Friday at ten.” This works across all templates and languages.

Voice Commands for Formatting

Simply say formatting commands out loud: “new paragraph”, “comma”, “bullet point” – and they are turned into real formatting. No manual cleanup, no interrupting your dictation.

Glossary for Project-Specific Terms

Technical terms, product names, your own class and function names: Verba can transcribe them exactly the way you want. Global terms go into settings.json, project-specific ones into a .verba-glossary.json in the project directory.

Offline Transcription with whisper.cpp

Anyone who wants maximum privacy or does not want to pay API costs for transcription can switch to whisper.cpp. The audio recording then never leaves the machine. Just install whisper.cpp, download a model, and set the provider in the settings to local.

Claude Code Integration

The new Claude Code Prompt template is designed for developers who use Claude Code. Verba transcribes the voice recording, enriches it with relevant codebase context (via grepai or OpenAI Embeddings) and generates an optimised prompt from it – ready to run in the terminal.

8 Templates – for Every Use Case

Verba ships with 8 ready-made templates:

Templates can be switched at any time with Cmd+Alt+T, or fully customised and extended in settings.json.

Cross-Platform: macOS, Windows and Linux

Verba runs everywhere VS Code runs. The audio recording uses platform-specific backends:

• macOS: AVFoundation
• Windows: DirectShow with automatic device detection
• Linux: PulseAudio

On every platform you can freely choose your microphone – either via the Quick Pick menu or via a setting.

Bring Your Own Key: No Subscription Costs

Verba is free. The extension is open source under the MIT licence.

Instead of a subscription, you use your own API keys:

• Deepgram API Key for Nova-3 transcription (or whisper.cpp for offline operation)
• Anthropic API Key for Claude text refinement

The keys are stored securely in VS Code’s SecretStorage – never in plain text, never in a config file. Anyone who transcribes locally only needs the Anthropic key.

What does this cost in practice? The API costs are typically a few cents per dictation. Verba tracks the costs automatically and shows an overview in a built-in WebView panel.

Verba vs. Commercial Alternatives

Getting Started

The prerequisite is ffmpeg for audio recording. On macOS a brew install ffmpeg, on Windows via Chocolatey or a direct download.

After that:

1. Install Verba from the VS Code Marketplace
2. Press Cmd+Shift+D
3. Enter API keys (one-time)
4. Choose a template – from the second time onwards, recording starts immediately

VS Code Marketplace: Verba

Verba is under active development. Feature requests and bug reports are welcome via GitHub Issues.