Licences
Polyloqui is built on open-source software, models and voices. Thank you to everyone who made them.
Speech recognition
- Whisper (tiny, base, small, large-v3-turbo) by OpenAI, MIT licence. Converted for on-device use by sherpa-onnx and onnx-community.
- Parakeet TDT 0.6B v3 by NVIDIA, Creative Commons Attribution 4.0 (CC BY 4.0).
- Silero VAD by Silero Team, MIT licence.
- TitaNet small by NVIDIA (NeMo), Apache License 2.0. In the app, optional: recognises the owner's voice and tells speakers apart.
- sherpa-onnx by the k2-fsa project (Next-gen Kaldi), Apache License 2.0.
Translation
- Google ML Kit on-device translation and language identification, under the Google APIs Terms of Service.
- Opus-MT models by Helsinki-NLP (University of Helsinki), CC BY 4.0, converted by Xenova (web version).
- M2M100 418M by Meta AI, MIT licence (web version, all-languages pack).
- Transformers.js by Hugging Face, Apache License 2.0; ONNX Runtime Web by Microsoft, MIT licence (web version).
Natural voices
Piper by Michael Hansen (rhasspy / Open Home Foundation), MIT licence, with these voices:
- Catalan, upc_ona: UPC, CC BY-SA 3.0 ES
- Czech, jirka: CC0
- Welsh, bu_tts: Techiaith, Bangor University, CC BY
- Danish, talesyntese: CC0
- German, thorsten: Thorsten Müller, CC0
- Greek, rapunzelina: CC0
- English, ljspeech: LJ Speech dataset (Keith Ito), public domain
- Spanish, sharvard: CC BY 3.0
- Basque, antton: CC BY 4.0
- Persian, reza_ibrahim: CC0
- Finnish, harri: CC0
- French, siwis: SIWIS database, CC BY 4.0
- Hungarian, anna: CC0
- Kazakh, issai: ISSAI, Nazarbayev University, CC BY 4.0
- Latvian, aivars: CC0
- Norwegian, talesyntese: CC0
- Nepali, google: Google, CC BY-SA 4.0
- Dutch, ronnie: CC0
- Polish, darkman: CC0
- Portuguese, faber: CC0
- Romanian, mihai: CC0
- Russian, dmitri: CC0
- Slovak, lili: CC0
- Slovenian, artur: CC BY 4.0
- Albanian, edon: CC0
- Swedish, nst: NST / National Library of Norway, CC0
- Ukrainian, ukrainian_tts: CC0
- Urdu, fasih: MIT licence
- Vietnamese, vais1000: CC BY 4.0
- Chinese, chaowen: CC0
Voice rooms
- RNNoise by Jean-Marc Valin / Xiph.Org, BSD licence; @sapphi-red/web-noise-suppressor, MIT licence.
- flutter_webrtc (Flutter WebRTC, MIT licence) and WebRTC (BSD licence).
- LiveKit client SDKs (livekit-client, livekit_client), Apache License 2.0.
- qrcode-generator by Kazuhiko Arase, MIT licence; jsQR, Apache License 2.0.
Hiding offensive words
- List of Dirty, Naughty, Obscene, and Otherwise Bad Words (LDNOOBW contributors), Creative Commons Attribution 4.0 International. Single characters in Chinese, Japanese, Korean and Thai are left out.
Design
- Bricolage Grotesque by Mathieu Triay, SIL Open Font License 1.1.
The Android app also lists the licences of every library it includes under Settings → Licences.