| Filename | Latest commit message | Latest commit date |
|---|---|---|
d9e0f89b89
Some checks failed
standards workspaces build + test / jouleclaw-rs (push) Successful in 3m1s
standards workspaces build + test / aac-rs (push) Successful in 1m24s
standards workspaces build + test / joule-code-rs (push) Successful in 54s
standards workspaces build + test / joulecontract-rs (push) Successful in 59s
no-vanity-metrics / Scan committed Markdown for vanity metrics (push) Successful in 4s
standards workspaces build + test / map-rs (push) Successful in 1m8s
standards workspaces build + test / mesh-rs (push) Successful in 40s
standards workspaces build + test / eoc-rs (push) Successful in 1m49s
standards workspaces build + test / jcp-rs (push) Successful in 1m25s
standards workspaces build + test / openpay (push) Successful in 3m24s
standards workspaces build + test / proof-rs (push) Successful in 1m16s
standards workspaces build + test / sandbox-rs (push) Successful in 2m9s
standards workspaces build + test / smart-byte-rs (push) Successful in 2m36s
standards workspaces build + test / arl-rs (push) Has been cancelled
standards workspaces build + test / joule-ui-rs (push) Has been cancelled
standards workspaces build + test / jouledesk-rs (push) Has been cancelled
standards workspaces build + test / joulehook-rs (push) Has been cancelled
standards workspaces build + test / sense-rs (push) Has been cancelled
standards workspaces build + test / macOS energy backend (jcp-energy) (push) Has been cancelled
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]> |
||
| .. | ||
| index.json | ||
| LICENSES.md | ||