Acknowledgements
Halm is built on the work of others. This page lists the third-party source, binaries, and on-device voice models that Halm vendors, bundles, ships, or downloads, together with their licenses.
For questions about any attribution here, email support@halm.app.
espeak-ng
Halm bundles espeak-ng (the eSpeak NG speech synthesizer's phonemization / espeak-ng-data text-analysis data) as the grapheme-to-phoneme front end for the on-device Piper neural voices.
- Project: espeak-ng — https://github.com/espeak-ng/espeak-ng
- License: GNU General Public License v3.0 (GPL-3.0-or-later) — https://www.gnu.org/licenses/gpl-3.0.html
Written offer of source. In accordance with GPL-3.0, the complete corresponding source code for the espeak-ng version used by this app is available at https://github.com/espeak-ng/espeak-ng. On request we will also provide the exact corresponding source for the build we ship; contact support@halm.app.
sherpa-onnx
Halm uses sherpa-onnx (k2-fsa) as the on-device neural-TTS inference runtime that drives the Piper VITS voices.
- Project: sherpa-onnx — https://github.com/k2-fsa/sherpa-onnx
- Copyright: © 2023-2025 Xiaomi Corporation (k2-fsa / next-gen Kaldi project)
- License: Apache License 2.0 — https://www.apache.org/licenses/LICENSE-2.0
ONNX Runtime
Halm links ONNX Runtime, the cross-platform ML inference engine that sherpa-onnx executes the voice models on.
- Project: ONNX Runtime — https://github.com/microsoft/onnxruntime
- Copyright: © Microsoft Corporation
- License: MIT License — https://github.com/microsoft/onnxruntime/blob/main/LICENSE
Piper voices — "Amy" and "Alan"
The downloadable neural voices Amy (en_US-amy-medium) and Alan (en_GB-alan-medium) are Piper VITS voice models from the Piper / rhasspy sample set.
- License: Creative Commons Attribution-ShareAlike 4.0 International (CC-BY-SA 4.0) — https://creativecommons.org/licenses/by-sa/4.0/
- Upstream: Piper — https://github.com/rhasspy/piper · Piper voices — https://github.com/rhasspy/piper-voices
Attribution & ShareAlike. These voice models are used under CC-BY-SA 4.0. Credit is given to the Piper / rhasspy project and the original voice-dataset authors. As required by the ShareAlike term, any adaptation or redistribution of these models is licensed under the same CC-BY-SA 4.0 license.
SwipeCardsKit
Part of Halm's swipe-card UI is adapted from SwipeCardsKit, vendored and reworked for Halm.
- Project: SwipeCardsKit — https://github.com/tobi404/SwipeCardsKit
- Copyright: © 2025 Beka Demuradze
- License: MIT License
Piper voices — Thorsten, Eva, Joe and Cori
Beyond Amy and Alan above, the shipped on-device voice roster consists of these Piper VITS voice models (upstream: Piper — https://github.com/rhasspy/piper · Piper voices — https://github.com/rhasspy/piper-voices):
- Thorsten (German) — from the *Thorsten-Voice* project, CC0-1.0 (public-domain dedication). No attribution is legally required; credit is given to Thorsten Müller and the Thorsten-Voice project as a courtesy. Covers the medium and HD tiers.
- Eva (
eva_k, German) — distributed under a BSD-3-Clause-style license, which requires attribution: credit to the eva_k voice-dataset author (M-AILABS / LibriVox lineage) and the Piper / rhasspy project. - Joe (English, US) — CC0-1.0 (public-domain dedication); credit given as a courtesy.
- Cori (English, UK) — public domain / CC0-1.0; credit given as a courtesy. Covers the medium and HD tiers.
Learning data sources
Halm's bundled learning data (word frequencies, CEFR level labels, example sentences, and the vocabulary calibration test) derives from these public sources:
- Leipzig Corpora Collection — German word frequencies derive from the downloadable corpus
deu_news_2022_1M(https://wortschatz.uni-leipzig.de/), used under CC BY 4.0. Citation: Goldhahn, Eckart & Quasthoff (2012), *Building Large Monolingual Dictionaries at the Leipzig Corpora Collection: From 100 to 200 Languages*, LREC 2012. - OpenSubtitles / OPUS — a spoken-register frequency signal derives from the OPUS OpenSubtitles v2018 corpora (https://opus.nlpl.eu/), built from http://www.opensubtitles.org/ data. Citation: Lison & Tiedemann (2016), *OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles*, LREC 2016.
- CEFR-J Wordlist v1.6 — English A1–B2 CEFR level labels (Tono Laboratory, Tokyo University of Foreign Studies, https://cefr-j.org/), used with acknowledgement per the CEFR-J terms, via the Open Language Profiles mirror (https://github.com/openlanguageprofiles/olp-en-cefrj).
- Octanove Goodman C1/C2 list — English C1/C2 CEFR level labels (Open Language Profiles, https://github.com/openlanguageprofiles), used under CC BY-SA 4.0; adaptations of these labels are shared under the same license.
- Tatoeba — example sentences from the Tatoeba Project (https://tatoeba.org/), used under CC BY 2.0 FR (with a CC0 subset).
- LexTALE-DE — the vocabulary-calibration methodology derives from
gasparl/lextale(https://github.com/gasparl/lextale), BSD-2-Clause.
Swift packages
Halm also uses the following open-source Swift packages, each under its own license (MIT unless noted):
- ConfettiSwiftUI (MIT) — https://github.com/simibac/ConfettiSwiftUI
- FeedKit (MIT) — https://github.com/nmdias/FeedKit
- IntelligenceGlow (MIT) — https://github.com/Livsy90/IntelligenceGlow
- Nuke (MIT) — https://github.com/kean/Nuke
- Pow (MIT) — https://github.com/movingparts-io/Pow
- swift-fsrs — https://github.com/open-spaced-repetition/swift-fsrs
- SwiftSoup (MIT) — https://github.com/scinfu/SwiftSoup
- SwiftUI-Shimmer (MIT) — https://github.com/markiv/SwiftUI-Shimmer