Linux desktop and Sailfish OS app for note taking, reading and translating with offline Speech to Text, Text to Speech and Machine Translation

• Description
• Languages and Models
• How to install
• Flatpak packages
• Beta version
• Building from sources
• How to enable a custom model
• Contributing to Speech Note
• How to support
• Reviews and demos
• License

Speech Note let you take, read and translate notes in multiple languages. It uses Speech to Text, Text to Speech and Machine Translation to do so. Text and voice processing take place entirely offline, locally on your computer, without using a network connection. Your privacy is always respected. No data is sent to the Internet.

Speech Note uses many different processing engines to do its job. Currently these are used:

• Speech to Text (STT)
  • Coqui STT (a fork of Mozilla DeepSpeech)
  • Vosk
  • whisper.cpp
  • Faster Whisper
  • april-asr
• Text to Speech (TTS)
  • espeak-ng
  • MBROLA
  • Piper
  • RHVoice
  • Coqui TTS
  • Mimic 3
  • WhisperSpeech
• Machine Translation (MT)
  • Bergamot Translator

Following languages are supported:

^{(e) experimental, most likely doesn't work well}

Faster Whisper, Coqui TTS and Mimic3 models are only available on x86-64.

Language models can be downloaded directly from the app.

Details of models which are currently configured for download are described in models.json (GitHub) or models.json (GitLab).

• Linux Desktop: Flatpak
• Arch Linux (AUR):
  • dsnote
  • dsnote-git
• Sailfish OS: OpenRepos

Starting from v4.4.0, the app distributed via Flatpak (published on Flathub) consists of the following packages:

• Base package "Speech Note" (net.mkiol.SpeechNote)
• Add-on for AMD graphics card "Speech Note AMD" (net.mkiol.SpeechNote.Addon.amd)
• Add-on for NVIDIA graphics card "Speech Note NVIDIA" (net.mkiol.SpeechNote.Addon.nvidia)

Base package includes all the dependencies needed to run every feature of the application. Add-ons add the capability of GPU acceleration, which speeds up some operations in the application.

Base package and add-ons contain many "heavy" libraries like CUDA, ROCm, Torch and Python libraries. Due to this, the size of the packages and the space required after installation are significant. If you don't need all the functionalities, you can use much smaller "Tiny" package (available on Releases page), which provides only the basic features. If you need, you can also use "Tiny" packages together with GPU acceleration add-on.

Comparison between Base, Tiny and Add-ons Flatpak packages:

In addition to the stable version in the Flathub repository, you can try to test the "Beta" version of the upcoming release. This version is usable, but may contain more bugs.

Beta version is available in "flathub-beta" repository. Follow these instructions to enable flathub-beta on your computer.

It is also possible to build and install the latest development (git) or latest stable (release) version from the repository using the provided PKGBUILD file (please note that the same remarks about building on Linux apply):

git clone <git repository url>

cd dsnote/arch/git      # build latest git version
# or
cd dsnote/arch/release  # build latest release version

makepkg -si

git clone <git repository url>

cd dsnote/flatpak

flatpak-builder --user --install-deps-from=flathub --repo="https://accionvegana.org/accio/0ITbvNmLiVHa0l2Z6MHc0/path/to/local/flatpak/repo" "https://accionvegana.org/accio/0ITbvNmLiVHa0l2Z6MHc0/path/to/output/dir" net.mkiol.SpeechNote.yaml

git clone <git repository url>

cd dsnote
mkdir build
cd build

sfdk config --session specfile=../sfos/harbour-dsnote.spec
sfdk config --session target=SailfishOS-4.4.0.58-aarch64
sfdk cmake ../ -DCMAKE_BUILD_TYPE=Release -DWITH_SFOS=ON -DWITH_PY=OFF
sfdk package

Speech Note has many build-time and run-time dependencies. This includes shared and static libraries, 3rd-party executables, Python and Perl scripts. Because of these complexity, the recommended way to build is to use Flatpak tool-chain (Flatpak manifest file and flatpak-builder). If you want to make a direct build (i.e. without flatpak) it is also possible but more complicated.

git clone <git repository url>

cd dsnote
mkdir build
cd build

cmake ../ -DCMAKE_BUILD_TYPE=Release -DWITH_DESKTOP=ON
make

To make build without support for Python components, add -DWITH_PY=OFF in cmake step.

To see other build options search for option(BUILD_XXX) in CMakeList.txt file.

All models available for download are specified in the configuration file (config/models.json). To enable a custom model that is compatible with currently supported engines, simply edit this file and restart the application.

When you first run the application, the models configuration file is created in:

• ~/.local/share/net.mkiol/dsnote/models.json, or
• ~/.var/app/net.mkiol.SpeechNote/data/net.mkiol/dsnote/models.json (Flatpak), or
• ~/.local/share/org.mkiol/dsnote/models.json (Sailfish OS)

You can freely edit currently enabled models or add new ones.

Model definition looks like this:

{
    "name": "<model name>",
    "model_id": "<model unique id>",
    "engine": "<engine type>",
    "lang_id": "<lang id>",
    "checksum": "<md5 checksum>",
    "checksum_quick": "<partial md5 checksum>",
    "comp": "<compression type",
    "urls": [
        <model URLs>
    ],
    "size": "<download size of all files>"
}

Allowed engine types: stt_ds, stt_vosk, stt_april, stt_whisper, stt_fasterwhisper, tts_piper, tts_rhvoice, tts_espeak, tts_coqui, tts_mimic3, mnt_bergamot

Allowed compression types: none, gz, xz, tarxz, targz, zip, zipall, dir, dirgz

Allowed URL types: http, https, file

Checksums are calculated for all files after unpacking. If you are adding a new model, you can use the --gen-checksums command line option to find the right checksums. To do this, put empty strings in both checksum and checksum_quick, save the file and run Speech Note with the mentioned option.

For example:

{
    "name": "New Piper Voice",
    "model_id": "en_piper_new",
    "engine": "tts_piper",
    "lang_id": "en",
    "checksum": "",
    "checksum_quick": "",
    "size": ""
    "comp": "dir",
    "urls": [
        "file:///home/me/models/new-model-medium.onnx",
        "file:///home/me/models/new-model-medium.onnx.json"
    ]
}

flatpak run net.mkiol.SpeechNote --verbose --gen-checksums

Any contribution is very welcome!

Project is hosted both on GitHub and GitLab. Feel free to make a PR/MR, report an issue or reqest for new feature on the platform you prefer the most.

Translation files in Qt format are in translations directory.

Preferred way to contribute translation is via Transifex service, but if you would like to make a direct PR/MR, please do it.

If you find Speech Note useful and would like to support this project, please consider doing one or two of the following:

• Give a ⭐ on GitHub or/and GitLab.
• Write a review in your applications manager app (Discover, Software or any other).
• Tell others about this app by mentioning it on social media.
• If you have spare money, make a small donation via Liberapay.

Speech Note relies on following open source projects:

• Qt
• Coqui STT
• Coqui TTS
• Vosk
• whisper.cpp
• WebRTC VAD
• libarchive
• RNNoise-nu
• {fmt}
• Hugging Face Transformers
• Piper
• RHVoice
• ssplit-cpp
• espeak-ng
• bergamot-translator
• Rubber Band Library
• simdjson
• Nlohmann JSON
• uroman
• astrunc
• FFmpeg
• LAME
• Vorbis
• TagLib
• libnumbertext
• KDBusAddons
• QHotkey
• faster-whisper
• Mimic 3
• Unikud
• april-asr
• Opus
• html2md
• maddy
• WhisperSpeech

• Speech Note 4.6 changes video (Speech Note 4.6)
• Screenshots (Speech Note 4.5)
• Speech Note 4.5 changes video (Speech Note 4.5)
• Marco's Box (Speech Note 4.4, Italian)
• Marco's Box video (Speech Note 4.4, Italian)
• alternativalinux (Speech Note 4.4, Italian)
• alternativalinux video (Speech Note 4.4, Italian)
• ZDNET (Speech Note 4.2)
• Translator feature video demo on Sailfish OS (Speech Note 4.0)
• Translator feature video demo on PinePhone (Speech Note 4.0)
• DebugPoint.com (Speech Note 4.0)
• DebugPoint.com video (Speech Note 4.0)
• OMG! Linux (Speech Note 4.0)
• LinuxLinks (Speech Note 4.0)
• The Linux Cast video (Speech Note 4.0)
• CONNECTwww.com (Speech Note 4.0)

Speech Note is an open source project. Source code is released under the Mozilla Public License Version 2.0.

3rd party libraries:

• Coqui STT, released under the Mozilla Public License Version 2.0
• Coqui TTS, released under the Mozilla Public License Version 2.0
• Vosk API, released uder the Apache License 2.0
• whisper.cpp, released under the MIT License
• WebRTC, released under this license
• libarchive, released under the BSD License
• RNNoise-nu, released under the BSD 3-Clause License
• {fmt}, released uder this license
• Hugging Face Transformers, released under the Apache License 2.0
• Piper, released under the MIT License
• RHVoice, released under the GNU General Public License v2.0
• ssplit-cpp, released under the Apache License 2.0
• espeak-ng, released under the GNU General Public License v3.0
• bergamot-translator, released under the Mozilla Public License 2.0
• Rubber Band Library, released under the GNU General Public License (version 2 or later)
• simdjson, released under the Apache License 2.0
• Nlohmann JSON, released under the MIT License
• uroman, released under this license
• astrunc, released under the MIT License
• FFmpeg, released under the GNU Lesser General Public License version 2.1 or later
• LAME, released under the LGPL
• Vorbis, released under this license
• TagLib, released under the GNU Lesser General Public License (LGPL) and Mozilla Public License (MPL)
• libnumbertext, released under the BSD License
• KDBusAddons, released under the LGPL licenses
• QHotkey, released under the BSD-3-Clause License
• faster-whisper, released under the MIT License
• Mimic 3, released under the AGPL-3.0 license
• Unikud, released under the MIT License
• april-asr, released under the GNU General Public License v3.0
• libopus, released under this license
• html2md, released under the MIT License
• maddy, released under the MIT License
• WhisperSpeech, released under the MIT License

The files in the directory nonbreaking_prefixes were copied from mosesdecoder project and distributed under the GNU Lesser General Public License v2.1.