Diktafon

   

Voice memos on cassette tapes — transcribed and summarised entirely on your device.

Diktafon is a voice-memo app built around one idea: capturing a thought and finding it again later must feel effortless. The feel is nostalgic — inspired by the tape devices of the past, memos live on topic-based cassettes, and each cassette plays as one continuous tape of chronologically ordered recordings.

Key features

• 📼 Simple to use — Controls you already know, inspired by real tape recorders: keep each topic on its own cassette — press record, talk, and rewind to find it again.
• 🚀 Modern — Automatic transcriptions and summaries: every memo transcribed, every cassette summarised and titled.
• 🔒 Completely private — Transcription and summaries are computed on your device; no accounts, no cloud, no analytics — nothing ever leaves it.
• 🌍 Built for you — Speaks your language, in recordings, summaries, and the app itself.
• ⓪ Free & open-source — No cost, no ads, no lock-in; MIT-licensed — read the code, build it yourself, make it yours.

_{English · Čeština · Deutsch · Español · Français · 한국어 · Polski · Português · Русский · Türkçe}

More features

• One-tap capture — open a cassette, press record. No naming, no setup, no network required.
• Tap-to-seek transcripts — tap any word in a transcript and the tape jumps right there.
• Summaries that keep up — a one-sentence gist of each longer memo, a summary of the whole cassette, and suggested titles for unnamed ones.
• Noise-aware transcription — voice-activity detection keeps background noise and silence from turning into invented sentences (no phantom words on a windy pause), and the default model adds a rumble filter that helps outdoor recordings. The original audio is never altered.
• The tape illusion — gapless playback with a colour-coded segment bar; memo boundaries are marked by a soft chime (optional). Scrub, seek, and skip across the whole tape as if it were one recording.
• Any language, any tape — the spoken language is auto-detected per memo, so one cassette may freely mix languages.
• Export & import — any cassette (or all of them) exports as a single archive with the audio and readable text inside; importing is additive — nothing gets overwritten.
• Little things — light & dark themes, cassette colour labels, retranscribe-cassette, per-memo copy and delete.
• On-device models — transcription by Whisper, summaries by Qwen. Models are downloaded once on first run (Wi-Fi recommended) and can be switched in Settings.

Install

Android

Get it on F-Droid or Google Play. Requires Android 7.0 (API 24) or newer; 4 GB RAM recommended for the default models. F-Droid distributes the same developer-signed APKs as the releases page, so F-Droid and sideloaded installs update interchangeably.

To sideload instead, download diktafon-<version>-android-arm64-v8a.apk (64-bit phones; the x86_64 APK is for emulators) from the releases page and install it — you may need to allow "install unknown apps" for your browser or file manager.

Alternatively, install from a computer with adb (USB debugging enabled on the device):

adb install diktafon-<version>-android-arm64-v8a.apk

iOS

Get it on the App Store. Requires iOS 13.0 or newer; a recent device is recommended for the default models.

Linux

Download diktafon-<version>-linux-x64.tar.gz from the releases page, then:

tar xzf diktafon-<version>-linux-x64.tar.gz
# Runtime dependencies: playback (libmpv), recording (parecord + ffmpeg)
sudo apt install libmpv2 pulseaudio-utils ffmpeg
./diktafon-<version>-linux-x64/diktafon

Built on Ubuntu 24.04 (glibc 2.39); older distributions should build from source instead.

Verifying a download

Each release ships a SHA256SUMS file, and all artifacts carry a GitHub build-provenance attestation proving they were built by this repository's release workflow:

sha256sum -c SHA256SUMS --ignore-missing
gh attestation verify diktafon-<version>-android-arm64-v8a.apk --repo jaromiru/diktafon

Building from source

Prerequisites:

• Flutter 3.44.5 (stable) — newer versions may work but are untested.
• Linux target: clang, cmake, ninja-build, pkg-config, libgtk-3-dev (plus the runtime packages above).
• Android target: Android SDK with NDK 28.2.13676358 and Java 17. whisper.cpp and llama.cpp are vendored under native/ and built automatically by the Flutter build (CMake / externalNativeBuild).

git clone https://github.com/jaromiru/diktafon.git
cd diktafon
flutter pub get

flutter run -d linux                     # desktop dev run
flutter build linux --release            # → build/linux/x64/release/bundle/
flutter build apk --release --split-per-abi \
  --target-platform android-arm64,android-x64  # → build/app/outputs/flutter-apk/

# Install on a connected Android device (-r replaces an existing install):
adb install -r build/app/outputs/flutter-apk/app-arm64-v8a-release.apk

Release APKs are signed with the debug key unless you provide android/key.properties (see the Flutter signing docs; the file and keystores are gitignored).

Tests

flutter analyze && flutter test          # static analysis + unit/widget tests
flutter test integration_test -d linux   # end-to-end flows on the desktop build

The integration tests cover record → play → transcribe → summarize flows. The transcription and summarization steps run the real engines only when pointed at locally downloaded models, and skip cleanly otherwise:

export DIKTAFON_WHISPER_MODEL=/path/to/ggml-tiny-q5_1.bin  # huggingface.co/ggerganov/whisper.cpp
export DIKTAFON_LLM_MODEL=/path/to/Qwen3-0.6B-Q8_0.gguf    # huggingface.co/Qwen/Qwen3-0.6B-GGUF
flutter test integration_test -d linux

Acknowledgments & License

Designed by Jaromír Janisch, implemented by Claude Code.

Licensed under MIT — see LICENCE.md for details, including the licences of the vendored engines, bundled fonts, the bundled Silero VAD model, and the runtime-downloaded models. The software is provided "as is", without warranty of any kind.