Mirent is built on open source software. This page lists the third-party components we use to generate speech and align words to audio, along with their licenses and the attribution their authors ask for.
Speech synthesis
VieNeu-TTS (Vietnamese voices)
Vietnamese speech in Mirent is generated by VieNeu-TTS v3-Turbo by Phan Nguyen Ngoc Bao (pnnbao97). Licensed under the Apache License 2.0, which covers the model weights, configs and the bundled preset voices.
The preset voices shipped with the model are used with the rights granted by their speakers, as stated by the model's authors.
MOSS-Audio-Tokenizer-Nano
The neural audio codec inside VieNeu-TTS, by the OpenMOSS Team. Licensed under the Apache License 2.0.
sea-g2p
Vietnamese and English grapheme-to-phoneme conversion, by pnnbao97. Licensed under the Apache License 2.0.
Kokoro (English voices)
English speech is generated by Kokoro-82M by hexgrad. Licensed under the Apache License 2.0.
Kokoro's training data included two Creative Commons datasets that ask for attribution: Koniwa (CC BY 3.0) and SIWIS (CC BY 4.0).
Misaki
English grapheme-to-phoneme conversion for Kokoro, by hexgrad. Licensed under the Apache License 2.0.
eSpeak NG
Used by Misaki to pronounce words outside its dictionary. eSpeak NG is licensed under the GNU General Public License v3.0. It runs as a separate program on our servers and is not distributed as part of the Mirent apps. Source is available from the eSpeak NG project.
Word timing and highlighting
wav2vec2-base-vietnamese
Aligns generated Vietnamese audio to the source text so words highlight as they are spoken. By dragonSwing, licensed under the Apache License 2.0. Built on Meta AI's wav2vec 2.0 architecture.
faster-whisper
A fallback aligner, based on OpenAI's Whisper and SYSTRAN's faster-whisper with CTranslate2. Licensed under the MIT License.
Machine learning frameworks
Mirent's audio pipeline also uses PyTorch and torchaudio (BSD-3-Clause), Hugging Face Transformers (Apache 2.0), librosa (ISC), and SoundFile (BSD-3-Clause).
Bundled in the iOS app
The software above runs on our servers. These libraries are compiled into the Mirent iOS app itself: Firebase SDK and the Google libraries it depends on — GoogleUtilities, AppCheck, GoogleDataTransport, GTMSessionFetcher, Promises, gRPC and Abseil (all Apache 2.0); LevelDB (BSD-3-Clause, Copyright © 2011 The LevelDB Authors); Nanopb (zlib, Copyright © 2011 Petteri Aimonen); RevenueCat (MIT); and Sentry (MIT).
Typefaces
The reading typeface in the app is Literata, Copyright © 2017 The Literata Project Authors. This website uses Geist. Both are licensed under the SIL Open Font License, Version 1.1.
License texts
Full license texts are available at the links above. Copies of the Apache License 2.0, the MIT License, the BSD-3-Clause License and the GNU GPL v3.0 are also included in the NOTICE file distributed with Mirent's source.
Questions about attribution or licensing: support@mirent.io.