Diktafon
github.com
- Free plan
- Windows
- macOS
- Linux
- iPhone and iPad
- Android
What Diktafon is
Diktafon is a free, open-source voice memo app for Android, iOS, Linux, and macOS. It records memos, transcribes them on the device with Whisper, and creates summaries with Qwen. Models download once at first launch and can be changed in Settings. Recordings are arranged in topic-based cassettes, with spoken language detected separately for each memo, allowing several languages in one cassette. Selecting a transcript word jumps playback to that position in the recording. Cassettes can be exported as archives containing audio and readable text, while imports add material without replacing existing items. The privacy policy says recordings, transcripts, summaries, labels, and settings remain in private app storage, with no data collection or sharing. Its only stated network activity is model downloads over HTTPS from Hugging Face and its CDN when requested. There are no accounts, analytics, crash reporting, advertising, or third-party service SDKs.
Who it suits
Diktafon suits people who want on-device transcription and summaries without accounts or advertising. It also fits users who organize multilingual recordings into exportable cassettes.
Good to know, and what to watch
- On-device Whisper transcription and Qwen summaries
- Transcript words jump to matching audio positions
- Cassettes support recordings in multiple languages
- Exports audio and readable text together
- No accounts, analytics, ads, or data sharing
- Android requires version 7.0 or newer
- Four GB RAM recommended for default Android models
- Linux setup lists several runtime dependencies
Plans
Every plan and priceFree and open-source
Free plan
The verdict
Diktafon combines voice recording, transcription, summaries, and cassette organization while keeping stated data inside private app storage. Its platform coverage and export options suit users who want local processing and portable archives.
What the maker publishes
- Accounts and analytics
- The app has no user accounts, analytics, crash reporting, advertising, or third-party service SDKs, according to its privacy policy. source
- Android requirements
- Android 7.0 (API 24) or newer is required, and 4 GB RAM is recommended for the default models. source
- Download verification
- Each release ships a SHA256SUMS file, and the repository says release artifacts include GitHub build-provenance attestations. source
- Export and import
- Cassettes can be exported as archives containing audio and readable text, and imports add content without overwriting existing items. source
- Headquarters
- Všenory, Czechia source
- iOS requirements
- The iOS app requires iOS 13.0 or newer, and the project recommends a recent device for the default models. source
- Languages
- The spoken language is detected for each memo, so a cassette can contain recordings in multiple languages. source
- Linux requirements
- The Linux instructions list libmpv, parecord, and ffmpeg as runtime dependencies, and say the build uses Ubuntu 24.04 with glibc 2.39. source
- Maker
- The project says it was designed by Jaromír Janisch. source
- Network use
- The policy says the app’s only network activity is downloading machine-learning models over HTTPS from Hugging Face and its CDN at the user’s request. source
- On-device privacy
- The privacy policy says recordings, transcripts, summaries, labels, and settings stay in the app’s private storage, and the app collects and shares no data. source
- Purpose
- Diktafon is a voice memo app that transcribes and summarizes recordings on the device and organizes them into topic-based cassettes. source
- Support
- The privacy policy lists [email protected] and opening a GitHub issue as contact options. source
- Transcription and summaries
- The app uses Whisper for transcription and Qwen for summaries, with models downloaded once on first run and switchable in Settings. source
- Transcript playback
- Tapping a word in a transcript jumps playback to that point in the recording. source
Checked against the maker's pages on 2 Oct 2026.