Back to browse
GitHub Repository

Diktafon is a private, open-source voice memo app that transcribes and summarizes your recordings entirely on your device — organized like cassette tapes.

3 starsC++

Diktafon – Voice memos on cassette tapes, transcribed on-device

by jaromiru·Jul 22, 2026·1 point·0 comments

AI Analysis

●●●BangerCozyDark HorseZero to One

Cassette tape UI makes on-device Whisper and Qwen summaries feel tactile and private.

Strengths
  • Silero VAD integration prevents silence from generating hallucinated transcript text.
  • Vendored C shims unify whisper.cpp and llama.cpp builds behind a single FFI layer.
  • Tap-to-seek transcript sync jumps audio playback to exact spoken word locations.
Weaknesses
  • Nostalgic cassette metaphor may confuse users expecting standard folder hierarchies.
  • Heavy local models require significant storage space on lower-end Android devices.
Category
Target Audience

Privacy-focused note-takers and offline-first mobile users

Similar To

WalkWrite · Notely Voice · Otter.ai

Post Description

Diktafon is a voice-memo app: each topic is a cassette, memos are appended to its tape, and a cassette plays back as one continuous recording. Every memo is transcribed and every cassette summarised, fully on-device: whisper.cpp for transcription, Qwen3 via llama.cpp for summaries — vendored behind two small C shims sharing one ggml build, driven from Dart (Flutter) over FFI in worker isolates. A bundled Silero VAD keeps silence and noise from turning into invented sentences. Models are downloaded once, sha256-pinned. No accounts, no cloud, no analytics. MIT; Android, iOS, Linux.

It came from a real need. The closest existing apps are WalkWrite (iOS-only, aimed at writers — speech becomes polished text) and Notely Voice (a notes app fed by dictation); both turn audio into text, while I wanted the recordings to stay the primary object, with the text attached.

Although I'm an experienced programmer, I wrote almost none of this by hand. I kept to the role of designer and reviewer, and Claude (Fable) wrote the code — an experiment in how far that gets. It did as I expected — very well, app complete in about a week of occasional 30 minutes sessions. Happy to answer questions about that workflow as much as the app.

I use it daily and I'm happy with it. I'll be glad for any feedback.

Source: https://github.com/jaromiru/diktafon

Similar Projects

AI/ML●●●Banger

I put Claude Code inside a Telegram bot for voice memos

Claude Agents browse your voice memo archive — actually novel bridge between transcription and agentic reasoning.

Solve My ProblemZero to One
buryhuang
104mo ago