Mumble Dictation: Local ASR with Personal Vocabulary

Most dictation tools are just glorified wrappers for cloud APIs that fail the moment you mention a niche library or a teammate's name. The "privacy" claims usually fall apart once you realize your audio is being streamed to a server for processing.

Mumble Dictation takes a different approach by keeping the entire ASR and cleanup pipeline on-device. The actual technical advantage here isn't just the local deployment, but how it handles the "vocabulary gap." It builds a local personal dictionary by analyzing what you type and what's on your screen. During the decoding phase, it boosts these specific terms, meaning technical jargon like "Supabase" or "PowerSync" actually gets spelled correctly instead of becoming a phonetic mess.

Beyond the local ASR, there are a few features that make it a viable AI workflow tool:

• Custom Transforms: You can set prompts to automatically turn transcripts into bullet points or concise summaries.
• LLM Flexibility: While the ASR is local, you can plug in your own API keys from OpenAI, Anthropic, or even Ollama if you prefer a specific model for the transcript cleanup phase.
• Voice Shortcuts: It can trigger actions, like searching Amazon or sending a prompt to Claude, via voice commands.

It supports 40 languages and avoids the subscription trap with a one-time purchase fee. For anyone needing a real-world deployment of local voice-to-text that doesn't sacrifice accuracy for privacy, this is a solid alternative to the standard OS dictation.

https://heymumble.com/dictation