Transaction-Science/open-standards
Watch
1
0
Fork
You've already forked open-standards
 0
Code Issues Pull requests Projects Releases 17 Packages Wiki Activity Actions 60
open-standards/aac/aac-rs/data/voices/LICENSES.md
Transaction Science d9e0f89b89
Some checks failed
standards workspaces build + test / jouleclaw-rs (push) Successful in 3m1s
Details
standards workspaces build + test / aac-rs (push) Successful in 1m24s
Details
standards workspaces build + test / joule-code-rs (push) Successful in 54s
Details
standards workspaces build + test / joulecontract-rs (push) Successful in 59s
Details
no-vanity-metrics / Scan committed Markdown for vanity metrics (push) Successful in 4s
Details
standards workspaces build + test / map-rs (push) Successful in 1m8s
Details
standards workspaces build + test / mesh-rs (push) Successful in 40s
Details
standards workspaces build + test / eoc-rs (push) Successful in 1m49s
Details
standards workspaces build + test / jcp-rs (push) Successful in 1m25s
Details
standards workspaces build + test / openpay (push) Successful in 3m24s
Details
standards workspaces build + test / proof-rs (push) Successful in 1m16s
Details
standards workspaces build + test / sandbox-rs (push) Successful in 2m9s
Details
standards workspaces build + test / smart-byte-rs (push) Successful in 2m36s
Details
standards workspaces build + test / arl-rs (push) Has been cancelled
Details
standards workspaces build + test / joule-ui-rs (push) Has been cancelled
Details
standards workspaces build + test / jouledesk-rs (push) Has been cancelled
Details
standards workspaces build + test / joulehook-rs (push) Has been cancelled
Details
standards workspaces build + test / sense-rs (push) Has been cancelled
Details
standards workspaces build + test / macOS energy backend (jcp-energy) (push) Has been cancelled
Details
feat(aac voices): hear a voice before downloading it; ratings measured 
A better voice is a 100 MB download, and the person could not hear
one before deciding. Making the samples showed something worse: some
voices did not say the words at all.

- Samples: each voice says a greeting and a request in its language
  (two or three seconds, MP3, about 10 KB). Settings > Better voices
  plays it with the sentence written beside it, before any download;
  checked by size and SHA-256 and played through the voices' own
  output. render_samples.mjs renders them with LAME; they are
  committed (Piper's synthesis is not repeatable), and build_voices.py
  publishes each beside the voices while it says the current text.
- Intelligibility: intelligibility.py has every voice say seven
  everyday phrases, a speech recognizer (Whisper large-v3-turbo on MLX)
  writes down what it hears, and the scores are committed. Ratings now
  come from that evidence (RATING_RULE): under 0.80 the voice is
  withheld; under 0.90, or any phrase under 0.60, experimental; 0.90
  and over clear; 0.95 and over with an upstream grade of A or B,
  natural.
- Withheld, because they say other words: German (the only German
  voice, 0.14), Dutch MLS (0.27), Ukrainian (0.18) and both Japanese
  voices (0.63-0.69: they misread 外, 好き and 疲れた). Their languages
  say so plainly under "Why". Portuguese and Hindi are experimental;
  Spanish, Chinese, Italian, Flemish and Cori are clear; Heart, Emma
  and Siwis natural. 16 voices in 8 languages, 452 MB of packs.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-30 20:20:25 -04:00

23 KiB
Raw Permalink Blame History

Neural voices: licences and attribution

Generated by tools/voices/build_voices.py for catalogue index.9f6a2c0577db.json (sources read 2026-09-29). Do not edit by hand.

Every neural voice runs entirely on the device, inside the browser; no text leaves it. Every one of them is a synthetic voice and should be labelled so wherever it is offered (this also meets the StyleTTS 2 pre-trained model condition that Kokoro inherits). Nothing below is non-commercial or GPL: all of it may be used in a free app and in a later paid tier. ShareAlike terms bind only the data or model files that carry them, not the app. The full licence and notice texts are in licences/, one file per runtime, component and voice.

Runtime

[email protected]

Shared components

kokoro-core@1

  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M v1.0 by hexgrad (Apache-2.0), built on StyleTTS 2 by Yinghao Aaron Li et al. Packed to int8 ORT stages by Moonshine AI. Its training data includes the SIWIS French Speech Synthesis Database (University of Edinburgh, CC BY 4.0) and Koniwa tnc (CC BY 3.0). All speech from this model is synthesized.
  • Size: 83.4 MB in 5 files
  • Full texts: licences/kokoro-core_1.da8cd789cd64.txt

g2p-en_us@1

  • Licence: BSD-2-Clause AND MIT (github.com/cmusphinx/cmudict@74790861f6/LICENSE)
  • Attribution: English pronunciations from the CMU Pronouncing Dictionary, Copyright (C) 1993-2015 Carnegie Mellon University. Out-of-vocabulary model and G2P by Moonshine AI (MIT).
  • Size: 25.0 MB in 4 files
  • Full texts: licences/g2p-en_us_1.cdebabb7433c.txt

g2p-fr@1

g2p-nl@1

  • Licence: CC-BY (github.com/open-dict-data/ipa-dict@43c3570eb3/README.md (credits))
  • Attribution: Dutch pronunciations provided by the Instituut voor de Nederlandse Taal (INT) under a CC BY licence, via ipa-dict by open-dict-data.
  • Note: ipa-dict states 'a CC BY license' for the INT data without a version; the CC BY 4.0 text is included as the current version for reference.
  • Size: 3.8 MB in 1 files
  • Full texts: licences/g2p-nl_1.9f0b95ca15c2.txt

g2p-hi@1

g2p-it@1

g2p-pt_br@1

g2p-zh_hans@1

  • Licence: CC-BY-3.0 AND Unicode-3.0 AND Apache-2.0 AND MIT (github.com/open-dict-data/ipa-dict@43c3570eb3/README.md (credits))
  • Attribution: Mandarin pronunciations from ipa-dict by open-dict-data (MIT), built from Unihan (Unicode licence), KFCD Pinyin (CC BY 3.0) and 開放漢語詞典 kaifangcidian.com (CC BY 3.0). Word segmentation: chinese-roberta-base-upos by Koichi Yasuoka (Apache-2.0), converted to ORT by Moonshine AI.
  • Size: 103.9 MB in 6 files
  • Full texts: licences/g2p-zh_hans_1.9d741fa61999.txt

Voices

en-kokoro-af_heart (en, default)

  • Voice: Heart (female, adult), kokoro; rated natural; quality: Kokoro grade A
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice af_heart by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-en_us@1; download 109.0 MB (122.3 MB with the runtime)
  • Full texts: licences/en-kokoro-af_heart.bf94172a2e9d.txt
  • Sample: samples/en/en-kokoro-af_heart.b66c9087db4c.mp3, the voice's own output (under its licence), saying "Hi! Could I have some more juice, please?"

en-kokoro-am_michael (en)

  • Voice: Michael (male, adult), kokoro; rated clear; quality: Kokoro grade C+
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice am_michael by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-en_us@1; download 109.0 MB (122.3 MB with the runtime)
  • Full texts: licences/en-kokoro-am_michael.84d5840eee55.txt
  • Sample: samples/en/en-kokoro-am_michael.f9f2c8961f42.mp3, the voice's own output (under its licence), saying "Hi! Could I have some more juice, please?"

en-kokoro-bf_emma (en)

  • Voice: Emma (British) (female, adult), kokoro; rated natural; quality: Kokoro grade B-
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice bf_emma by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-en_us@1; download 109.0 MB (122.3 MB with the runtime)
  • Full texts: licences/en-kokoro-bf_emma.0a9e68a8e3b2.txt
  • Sample: samples/en/en-kokoro-bf_emma.e687d39d8823.mp3, the voice's own output (under its licence), saying "Hi! Could I have some more juice, please?"

en-piper-cori (en)

  • Voice: Cori (British) (female, adult), piper; rated clear; quality: Piper high, British English; light and fast, clean on single words in testing
  • Licence: MIT (huggingface.co/rhasspy/piper-voices@c10ece1aad/en/en_GB/cori/high/MODEL_CARD)
  • Attribution: Piper voice en_GB-cori-high by Bryce Beattie (rhasspy/piper-voices, MIT), trained from scratch on LibriVox recordings (public domain). Converted to ORT format with onnxruntime 1.23.2.
  • Requires: g2p-en_us@1; download 140.0 MB (153.3 MB with the runtime)
  • Full texts: licences/en-piper-cori.5c98b20d0dd6.txt
  • Sample: samples/en/en-piper-cori.16165db3d777.mp3, the voice's own output (under its licence), saying "Hi! Could I have some more juice, please?"

es-kokoro-ef_dora (es, default)

  • Voice: Dora (female, adult), kokoro; rated clear; quality: Kokoro, ungraded upstream
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice ef_dora by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1; download 83.9 MB (97.3 MB with the runtime)
  • Full texts: licences/es-kokoro-ef_dora.888aaac15fd4.txt
  • Sample: samples/es/es-kokoro-ef_dora.c04700d37d5f.mp3, the voice's own output (under its licence), saying "¡Hola! Quiero más jugo, por favor."

es-kokoro-em_alex (es)

  • Voice: Alex (male, adult), kokoro; rated clear; quality: Kokoro, ungraded upstream
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice em_alex by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1; download 83.9 MB (97.3 MB with the runtime)
  • Full texts: licences/es-kokoro-em_alex.b4f19c0532c2.txt
  • Sample: samples/es/es-kokoro-em_alex.8bf7b811ec28.mp3, the voice's own output (under its licence), saying "¡Hola! Quiero más jugo, por favor."

fr-kokoro-ff_siwis (fr, default)

  • Voice: Siwis (female, adult), kokoro; rated natural; quality: Kokoro grade B-
  • Licence: Apache-2.0 AND CC-BY-4.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice ff_siwis by hexgrad (Apache-2.0). Voice trained on the SIWIS French Speech Synthesis Database (University of Edinburgh DataShare, https://datashare.ed.ac.uk/handle/10283/2353), CC BY 4.0.
  • Requires: kokoro-core@1, g2p-fr@1; download 115.4 MB (128.7 MB with the runtime)
  • Full texts: licences/fr-kokoro-ff_siwis.9c0075e52378.txt
  • Sample: samples/fr/fr-kokoro-ff_siwis.b52891911edf.mp3, the voice's own output (under its licence), saying "Bonjour ! Je voudrais encore un peu de jus, s'il vous plaît."

pt-kokoro-pf_dora (pt, default)

  • Voice: Dora (female, adult), kokoro; rated experimental; quality: Kokoro, ungraded upstream
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice pf_dora by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-pt_br@1; download 86.6 MB (99.9 MB with the runtime)
  • Full texts: licences/pt-kokoro-pf_dora.e0dd7148c87d.txt
  • Sample: samples/pt/pt-kokoro-pf_dora.6506192f2790.mp3, the voice's own output (under its licence), saying "Olá! Eu queria mais suco, por favor."

pt-kokoro-pm_alex (pt)

  • Voice: Alex (male, adult), kokoro; rated experimental; quality: Kokoro, ungraded upstream
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice pm_alex by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-pt_br@1; download 86.6 MB (99.9 MB with the runtime)
  • Full texts: licences/pt-kokoro-pm_alex.61809fd1a2a7.txt
  • Sample: samples/pt/pt-kokoro-pm_alex.de8be0ebeb9b.mp3, the voice's own output (under its licence), saying "Olá! Eu queria mais suco, por favor."

it-kokoro-if_sara (it, default)

  • Voice: Sara (female, adult), kokoro; rated clear; quality: Kokoro grade C
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice if_sara by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-it@1; download 85.7 MB (99.0 MB with the runtime)
  • Full texts: licences/it-kokoro-if_sara.4bfbcc9b16d0.txt
  • Sample: samples/it/it-kokoro-if_sara.9c187a5a095c.mp3, the voice's own output (under its licence), saying "Ciao! Vorrei ancora un po' di succo, per favore."

it-kokoro-im_nicola (it)

  • Voice: Nicola (male, adult), kokoro; rated clear; quality: Kokoro grade C
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice im_nicola by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-it@1; download 85.7 MB (99.0 MB with the runtime)
  • Full texts: licences/it-kokoro-im_nicola.7dcf87d63550.txt
  • Sample: samples/it/it-kokoro-im_nicola.e4292e40fd83.mp3, the voice's own output (under its licence), saying "Ciao! Vorrei ancora un po' di succo, per favore."

nl-piper-rdh (nl, default)

  • Voice: Reader (Flemish) (male, adult), piper; rated clear; quality: Piper medium, Belgian Dutch
  • Licence: MIT AND CC0-1.0 (huggingface.co/rhasspy/piper-voices@c10ece1aad/nl/nl_BE/rdh/medium/MODEL_CARD)
  • Attribution: Piper voice nl_BE-rdh-medium (rhasspy/piper-voices, MIT), trained from scratch on dutch-vl-tts by r-dh (CC0 1.0). Converted to ORT format with onnxruntime 1.23.2.
  • Requires: g2p-nl@1; download 67.4 MB (80.8 MB with the runtime)
  • Full texts: licences/nl-piper-rdh.4283bab2866d.txt
  • Sample: samples/nl/nl-piper-rdh.0c5eca7eff94.mp3, the voice's own output (under its licence), saying "Hallo! Mag ik nog wat sap, alsjeblieft?"

zh-kokoro-zf_xiaobei (zh, default)

  • Voice: Xiaobei (female, adult), kokoro; rated clear; quality: Kokoro grade D
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice zf_xiaobei by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-zh_hans@1; download 187.8 MB (201.1 MB with the runtime)
  • Full texts: licences/zh-kokoro-zf_xiaobei.b394de311c3a.txt
  • Sample: samples/zh/zh-kokoro-zf_xiaobei.e049569f81b2.mp3, the voice's own output (under its licence), saying "你好！请再给我一些果汁，谢谢。"

zh-kokoro-zm_yunjian (zh)

  • Voice: Yunjian (male, adult), kokoro; rated clear; quality: Kokoro grade D
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice zm_yunjian by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-zh_hans@1; download 187.8 MB (201.1 MB with the runtime)
  • Full texts: licences/zh-kokoro-zm_yunjian.04ee7cc75a78.txt
  • Sample: samples/zh/zh-kokoro-zm_yunjian.e90e6b133b91.mp3, the voice's own output (under its licence), saying "你好！请再给我一些果汁，谢谢。"

hi-kokoro-hf_alpha (hi, default)

  • Voice: Alpha (female, adult), kokoro; rated experimental; quality: Kokoro grade C
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice hf_alpha by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-hi@1; download 84.0 MB (97.3 MB with the runtime)
  • Full texts: licences/hi-kokoro-hf_alpha.a378377d7230.txt
  • Sample: samples/hi/hi-kokoro-hf_alpha.ac4c01e098e4.mp3, the voice's own output (under its licence), saying "नमस्ते! कृपया मुझे और जूस दीजिए।"

hi-kokoro-hm_omega (hi)

  • Voice: Omega (male, adult), kokoro; rated experimental; quality: Kokoro grade C
  • Licence: Apache-2.0 (huggingface.co/hexgrad/Kokoro-82M@f3ff357179/README.md)
  • Attribution: Kokoro-82M voice hm_omega by hexgrad (Apache-2.0).
  • Requires: kokoro-core@1, g2p-hi@1; download 84.0 MB (97.3 MB with the runtime)
  • Full texts: licences/hi-kokoro-hm_omega.f425c24f2af2.txt
  • Sample: samples/hi/hi-kokoro-hm_omega.9d20d970da64.mp3, the voice's own output (under its licence), saying "नमस्ते! कृपया मुझे और जूस दीजिए।"

Voices not offered

These pass the licence review but are not built or offered, because a voice must say what the person chose.

voice language reason
en-piper-norman (Norman) en The single word 'more' comes out as 3-5 s of babble, also under reference inference of the upstream model. Cori is offered instead.
de-piper-mls (Reader) de It says other words: in testing, a speech recognizer understood 14% of 7 everyday phrases it said.
nl-piper-mls (Reader (Netherlands)) nl It says other words: in testing, a speech recognizer understood 27% of 7 everyday phrases it said.
ja-kokoro-jf_alpha (Alpha) ja It says other words: in testing, a speech recognizer understood 69% of 7 everyday phrases it said.
ja-kokoro-jm_kumo (Kumo) ja It says other words: in testing, a speech recognizer understood 63% of 7 everyday phrases it said.
uk-piper-ukrainian_tts (Lada) uk It says other words: in testing, a speech recognizer understood 18% of 7 everyday phrases it said.

Languages without a neural voice

code status reason
de web-speech-only The voices with a licence this app can use (Reader) do not yet say the words reliably with this voice engine: in testing, a speech recognizer understood 14% of everyday phrases they said.
ja web-speech-only The voices with a licence this app can use (Alpha, Kumo) do not yet say the words reliably with this voice engine: in testing, a speech recognizer understood 63% to 69% of everyday phrases they said.
ko web-speech-only No voice with a verifiable licence: Piper ko_KR-kss is CC BY-NC-SA, and Moonshine's ko_KR-melotts-medium states no base checkpoint and its training code is not public.
ar web-speech-only Piper ar_JO-kareem is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, and its dataset licence is unstated. OpenBible Arabic is a later candidate pending a rights check.
he web-speech-only Piper he_IL-saspeech has an unverified dataset licence and is built on LibriTTS-R, which Piper's own card says is fine-tuned from lessac; also there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
ur web-speech-only Piper ur_PK-fasih is built on hi_IN-rohan, which is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis; aegis_female states no base; and there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
fa web-speech-only Every Persian Piper voice is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis or chained to one (amir, reza_ibrahim, ganji, ganji_adabi); gyro's licence and base are unclear; and there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
bn web-speech-only Piper bn_BD-google has CC BY-SA data but is built on LibriTTS-R, which Piper's own card says is fine-tuned from lessac; there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL). Rasa-13 and OpenBible Bengali await a rights review.
ta web-speech-only No verifiable voice: the AI4Bharat Rasa-13 card is gated, and the Open.Bible Tamil source rights could not be checked.
te web-speech-only Piper padmavathi and venkatesh (CC BY 4.0 data) state no base checkpoint, maya is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, and there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
ml web-speech-only Piper arjun and meera are built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis with an unstated dataset licence; Rasa-13 is gated and unverifiable.
mr web-speech-only Piper mr_IN-google is built on LibriTTS-R, which Piper's own card says is fine-tuned from lessac, and there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL). OpenBible Marathi awaits a rights review.
kn web-speech-only No verifiable voice: Rasa-13 is gated and the Open.Bible Kannada source rights are unverified.
gu web-speech-only No verifiable voice: the Open.Bible Gujarati source rights are unverified.
pa web-speech-only No verifiable voice: Rasa-13 is gated and the Open.Bible Punjabi source rights are unverified.
th web-speech-only Piper th_TH-tsync2 is CC BY-NC-SA 3.0 (non-commercial).
vi web-speech-only Piper vais1000 is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, vivos is CC BY-NC-SA, and 25hours_single is built on the Ryan voice (RyanSpeech, CC BY-NC-SA) with an unknown licence.
id web-speech-only Piper id_ID-news_tts is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, with an unstated dataset licence.
ms web-speech-only No Malay voice exists in Piper, Kokoro or Pocket TTS.
tl web-speech-only No Tagalog voice exists in Piper, Kokoro or Pocket TTS (OpenBible has Hiligaynon, not Tagalog).
ru web-speech-only Piper denis and dmitri are built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, irina's licence is unknown, and ruslan is CC BY-NC-SA.
pl web-speech-only Every Polish Piper voice is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis or built on the Ryan voice (RyanSpeech, CC BY-NC-SA).
uk web-speech-only The voices with a licence this app can use (Lada) do not yet say the words reliably with this voice engine: in testing, a speech recognizer understood 18% of everyday phrases they said.
ro web-speech-only Piper ro_RO-mihai is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis.
cs web-speech-only Piper jirka is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis; kasandra (CC BY 4.0) states no base checkpoint; and there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
sv web-speech-only Piper sv_SE-nst-medium is clean (CC0, trained from scratch), but there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
da web-speech-only Piper da_DK-talesyntese is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis.
nb web-speech-only Piper no_NO-nvcc-medium is clean (CC0, trained from scratch), but there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
fi web-speech-only Piper harri-medium is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis and harri-low is built on the Ryan voice (RyanSpeech, CC BY-NC-SA).
sw web-speech-only Piper sw_CD-lanfrica is built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis, with an unstated dataset licence. OpenBible Swahili awaits a rights review.
am web-speech-only No voice with a licence that allows commercial use (none in Piper; MMS-TTS is CC BY-NC).
ha planned OpenBible VITS Hausa (CC BY-SA 4.0, trained from scratch on Open.Bible audio) is verified; it needs an ONNX export and the onnxruntime-web runtime, planned for v1.1.
yo planned OpenBible VITS Yoruba (CC BY-SA 4.0, trained from scratch on Open.Bible audio) is verified; it needs an ONNX export and the onnxruntime-web runtime, planned for v1.1. Yoruba text must carry full tone marks.
zu web-speech-only No voice with a licence that allows commercial use (OpenBible has Ndebele, not Zulu; MMS-TTS is CC BY-NC).
tr web-speech-only Piper tr_TR-dfki is CC BY-NC-SA and built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis; fahrettin and fettah are not in the current piper-voices catalogue, so their licences cannot be checked.
el web-speech-only Piper el_GR-rapunzelina-medium is clean (CC0, trained from scratch), but there is no non-GPL G2P for it in the runtime yet (eSpeak NG is GPL).
hu web-speech-only Piper anna, berta and imre are all built on the lessac voice, whose Blizzard-2013 licence bars commercial synthesis.