# Audio credits

All Hebrew letter, number and color pronunciation recordings are from Wikimedia Commons.

Source files downloaded from Commons are `.wav`; they are re-encoded to mono
64 kbps `.mp3` before being committed here (see AGENTS.md).

## Letters (`he/*.mp3`)
- Most letters: recorded by **Uziel302**, via the [Lingua Libre](https://lingualibre.org) project.
  License: [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/) (public domain, no attribution required).
- **ו** (`vav.mp3`): Uziel302 never recorded "vav", so this is
  [`File:LL-Q9288 (heb)-YaronSh-וו.wav`](https://commons.wikimedia.org/wiki/File:LL-Q9288_(heb)-YaronSh-%D7%95%D7%95.wav)
  by **YaronSh**, lowered 4 dB to match the other letters. License: CC0 1.0.
  It replaced Buffer's `ו` (CC BY-SA 4.0), whose voice stood out more.
- **ת** (`tav.mp3`) is said **"taf"**, the everyday Israeli pronunciation. Uziel302's
  own `ת` and `תו` both say "tav", so the clip is the second syllable cut out of
  [`File:LL-Q9288 (heb)-Uziel302-נטף.wav`](https://commons.wikimedia.org/wiki/File:LL-Q9288_(heb)-Uziel302-%D7%A0%D7%98%D7%A3.wav)
  ("nataf") — `0.43s`–`0.82s`, i.e. from inside the ט closure through the ף —
  lowered 4 dB and padded to 250 ms of leading silence, 0.88 s total, like the old
  clip. Same speaker, but a different recording session, so it sounds slightly different.
  License: CC0 1.0.
- **ל** (`lamed.mp3`) is Uziel302's `ל`. It is louder and longer than the other letters.
  His `למד` recording is at a normal level, but it says "lamad" (the verb), not the letter name, so it is not used.
- **ב** (`bet.mp3`) is the odd one out: Lingua Libre has no standalone recording of the
  letter name בֵּית. The clip is the first word cut out of
  [`File:LL-Q9288 (heb)-Uziel302-בית קברות.wav`](https://commons.wikimedia.org/wiki/File:LL-Q9288_(heb)-Uziel302-%D7%91%D7%99%D7%AA_%D7%A7%D7%91%D7%A8%D7%95%D7%AA.wav)
  ("beit kvarot") — `0.095s`–`0.485s`, then:
  - slowed to `rubberband=tempo=0.80` (pitch-preserving time-stretch). In the
    source it's a construct-state word inside a phrase, so it's spoken faster
    than a citation-form word; at 0.80 its 0.39 s matches the comparable
    one-syllable letter names (shin 0.38 s, vav 0.37 s).
  - gain-matched, and padded to 250 ms of leading / ~330 ms of trailing silence
    so it has the same shape as the other letters (0.97 s total, speech at
    0.28–0.67 s, like dalet/ayin/nun/resh). Without that padding it fired
    instantly on tap and cut off dead, which reads as clipped and hurried even
    though the word itself is a normal length.
  Same speaker (Uziel302) as most of the alphabet, so the voice matches.
  License: CC0 1.0 (public domain, no attribution required).
  It previously used a **woman's** voice while every other letter is a man's, which
  is why it was replaced.

## Numbers (`he/numbers/*.mp3`)
- **1–7** (`1.mp3`–`7.mp3`): recorded by **Guy Lederfein** (Wikimedia Commons user "Dirk Gently"), 2006.
  Sources: `File:He-il-akhat.ogg`, `-shtaim`, `-shalosh`, `-arba`, `-khamesh`, `-shesh`, `-sheva` on Wikimedia Commons.
  License: [CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/) — attribution required.
- **8, 9** (`8.mp3`, `9.mp3`): recorded by **Uziel302** via Lingua Libre.
  License: CC0 1.0 (public domain, no attribution required).
- **10** (`10.mp3`, "עשר"): recorded by **Buffer** via Lingua Libre.
  Source: [`File:LL-Q9288 (heb)-Buffer-עשר.wav`](https://commons.wikimedia.org/wiki/File:LL-Q9288_(heb)-Buffer-%D7%A2%D7%A9%D7%A8.wav) on Wikimedia Commons.
  License: [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) — **attribution required**.
- **11–19, 30–90** (`11.mp3`–`19.mp3`, `30.mp3`–`90.mp3`): recorded by **YaronSh** via Lingua Libre.
  License: CC0 1.0 (public domain, no attribution required).
- **20, 40, 60** (`20.mp3`, `40.mp3`, `60.mp3`): recorded by **Uziel302** via Lingua Libre.
  License: CC0 1.0 (public domain, no attribution required).
- **100** (`100.mp3`, "מאה"): recorded by **Buffer** via Lingua Libre.
  Source: [`File:LL-Q9288 (heb)-Buffer-מאה.wav`](https://commons.wikimedia.org/wiki/File:LL-Q9288_(heb)-Buffer-%D7%9E%D7%90%D7%94.wav) on Wikimedia Commons.
  License: [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) — **attribution required**.
- **15** is the odd one out: Lingua Libre has no feminine `חמש עשרה`, so `15.mp3` is
  the masculine `חמישה עשר`. Every other clip 1–19 is the feminine counting form.
- **21–29, 31–39, … 91–99**: not pre-recorded. Hebrew says these as two words
  ("עשרים ואחת"), and no single-word recording exists. They stay silent unless the
  device has a real Hebrew TTS voice installed — the app never reads a Hebrew
  number in another language.

## Colors (`he/colors/*.mp3`)

All eleven are Lingua Libre recordings of the single color word, downloaded as
`.wav` from Wikimedia Commons and re-encoded here. Two speakers, because
neither one covers all eleven words — Lingua Libre has exactly one Hebrew
recording per word, so there is nothing to choose between:

- **אדום, צהוב, ירוק, כחול, חום, שחור, לבן** (`red`, `yellow`, `green`, `blue`,
  `brown`, `black`, `white`): recorded by **Buffer** via Lingua Libre.
  License: [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/) — **attribution required**.
  Same speaker as the numbers עשר and מאה.
- **כתום, סגול, ורוד, אפור** (`orange`, `purple`, `pink`, `gray`): recorded by
  **YaronSh** via Lingua Libre.
  License: [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/) (public domain, no attribution required).
  Same speaker as the number teens and round tens.

Source files follow the Commons naming pattern
`File:LL-Q9288 (heb)-<speaker>-<word>.wav`.

There are no English recordings: English `speechSynthesis` voices are installed
essentially everywhere, so the colors game speaks English through TTS and plays
these clips for Hebrew.