Voice credits
Hear Church speaks with voices other people recorded and shared. Thank you to them. The names shown in the app are our own labels and are not the names of the people recorded. Each voice was trained from the source below and is used unchanged, apart from adjusting its speed and volume.
| Language | Voice | Source | Licence |
|---|---|---|---|
| French | Male 1 - Julien | French voice “tom” by Tjiho | AGPLv3 |
| French | Female 1 - Camille | The SIWIS French Speech Synthesis Database, by Junichi Yamagishi, Pierre-Edouard Honnet, Philip Garner and Alexandros Lazaridis, fine-tuned from a model trained on the Hi-Fi TTS dataset (reader 92, Cori Samuel), by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (NVIDIA) | CC BY 4.0; Hi-Fi TTS: CC BY 4.0 |
| Wolof | Male 1 - Moussa | SpeechT5 Wolof, by Bilal Faye | MIT |
| Wolof | Female 1 - Fatou | SpeechT5 Wolof, by Bilal Faye | MIT |
| Spanish | Male 1 - Carlos | Ald Mexican Spanish speech dataset | Unlicense |
| Spanish | Female 1 - Lucía | Piper TTS Spanish, by HirCoir | Apache 2.0 |
| Russian | Male 1 - Ivan | Open Home Foundation voice datasets (“denis”) | CC0 |
| Russian | Male 2 - Alexei | Open Home Foundation voice datasets (“dmitri”), fine-tuned from a model trained on the Hi-Fi TTS dataset (reader 92, Cori Samuel), by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (NVIDIA) | CC0; Hi-Fi TTS: CC BY 4.0 |
| Portuguese | Male 1 - João | Open Home Foundation voice datasets (“cadu”) | CC0 |
| Portuguese | Male 2 - Pedro | Open Home Foundation voice datasets (“faber”), fine-tuned from a model trained on the Hi-Fi TTS dataset (reader 92, Cori Samuel), by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (NVIDIA) | CC0; Hi-Fi TTS: CC BY 4.0 |
| Ukrainian | Female 1 - Olena | Ukrainian TTS datasets (“tetiana”), voice by Roman Stasyshyn | Apache 2.0 |
| Ukrainian | Male 1 - Andriy | Ukrainian TTS datasets (“oleksa”), voice by Roman Stasyshyn | Apache 2.0 |
| Persian | Male 1 - Ali | Datacula TTS databases (“amir”) | CC0 |
| Persian | Male 2 - Reza | Datacula TTS databases (“ganji”) | CC0 |
| Nepali | Male 1 - Ram | Open Home Foundation voice datasets (“chitwan”) | CC0 |
| Vietnamese | Female 1 - Lan | VAIS-1000: A Vietnamese Speech Synthesis Corpus, submitted by Truong Do | CC BY 4.0 |
| Chinese | Female 1 - Mei | Open Home Foundation voice datasets (“chaowen”), fine-tuned from a model trained on the Hi-Fi TTS dataset (reader 92, Cori Samuel), by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (NVIDIA) | CC0; Hi-Fi TTS: CC BY 4.0 |
| English | Male 1 - James | Open Home Foundation voice datasets (“mike”) | CC0 |
| English | Female 1 - Mary | Open Home Foundation voice datasets (“kathleen”), fine-tuned from a model trained on the Hi-Fi TTS dataset (reader 92, Cori Samuel), by Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg and Yang Zhang (NVIDIA) | CC0; Hi-Fi TTS: CC BY 4.0 |
Except for Wolof, these voices come from the Piper voices collection, and speech is generated with Piper and eSpeak NG (GPL-3.0). Wolof uses Microsoft's SpeechT5 vocoder (MIT).
Wolof speech recognition
When someone speaks Wolof, the app understands it with models other people trained and shared. Thank you to them.
- On Macs with Apple silicon: Kiriku-Wolof-ASR, by AI Hub Senegal (Apache 2.0), built on OpenAI's Whisper (MIT). It learned from recordings that include the Kallaama speech corpus and Google FLEURS (both CC BY 4.0).
- On Intel Macs: Omnilingual ASR, by Meta (Apache 2.0).
Open-source software
The app includes open-source software under the GNU GPL and LGPL, among others: Piper and eSpeak NG (GPL v3), LAME and libsoxr (LGPL), and in the Intel version FFmpeg (LGPL v3) with x264 and x265 (GPL v2). Their licences and notices are in the app (Settings → About and help → Open-source notices). Their complete source code, for every version of the app, is at hearchurch.com/source; for at least three years after we last distribute a version, you may also ask support@hearchurch.com for it on a physical medium, for no more than our cost of sending it. You may replace the LGPL libraries in the app with modified versions; macOS then needs the app re-signed on your Mac: codesign --force --deep -s - "/Applications/Hear Church.app".