Credits
Polly the parrot's voice
Polly's recorded lines (public/audio/voice/parrot_*.ogg) are speech synthesised with Fish Audio Drama 3 using a real talking parrot as the reference voice:
- "Talking Parrot (Psittacula krameri).ogg" by Amada44, Wikimedia Commons, licensed CC BY 3.0 (https://creativecommons.org/licenses/by/3.0/), https://commons.wikimedia.org/wiki/File:Talking_Parrot_(Psittacula_krameri).ogg (read 2026-10-20). 12 seconds of a rose-ringed parakeet saying "I'm pretty quick!" and "Hello?".
The reference clip is kept in the sound lab, not in this repository (pirate-sounds/parrot/real/_reference_krameri_talking_CC-BY-3.0_Amada44.wav). Whether a voice cloned from a CC BY recording counts as an adaptation of it, and so what the attribution must say, has not been settled by a lawyer (unverified); the credit above is given in any case. The earlier pitched-up takes (not from a real bird) are kept in the lab: pirate-sounds/parrot/parrot.before-real-parrot.json.
Two other real parrot recordings were looked at and not used: "Psittacus erithacus speaking.oga" (Geoprofi Lars, CC BY-SA 4.0, says "Hello?") and "African grey parrot sound.wav" (Ganesh Mohan T, CC BY-SA 4.0, no words). ShareAlike would have bound the game's own audio.
Crew voices
The 42 crew lines (public/audio/voice/{steering,sails,anchor,cannons,patching,bailing,enemy_sighted,damage,low_health,treasure,island,storm,victory,idle}_{1,2,3}.ogg) are speech synthesised with Fish Audio Drama 3 (the Fish Audio Community Licence; read it before shipping commercially: unverified), using two real human reference voices performing genuine pirate characters instead of the generic Cartesia-TTS "clint"/"ben" voices used before 2026-10-21:
- Long John Silver, read by Barry Eads, from LibriVox's "Treasure Island (version 3, dramatic reading)" (https://librivox.org/treasure-island-dramatic-reading-by-robert-louis-stevenson/, read 2026-10-21). LibriVox recordings are Public Domain in the USA. The reference clip is chapter 11, "What I Heard in the Apple-Barrel" (https://archive.org/download/treasure_island_dram_1306_librivox/treasureisland_11_stevenson_64kb.mp3), 26.2-46.4s, Silver's own monologue about Flint's crew ("Flint was cap'n; I was quartermaster, along of my timber leg...").
- Israel Hands, read by Mark F. Smith, same LibriVox recording. The reference clip is chapter 26, "Israel Hands" (https://archive.org/download/treasure_island_dram_1306_librivox/treasureisland_26_stevenson_64kb.mp3), 414.4-434.3s, Hands's own dialogue ("I never seen good come o' goodness yet. Him as strikes first is my fancy. Dead men don't bite...").
Both characters are voiced by different named LibriVox volunteers, giving two genuinely distinct voices (not one narrator performing two characters), reading public-domain dialogue with real "weathered sea dog" dialect and delivery. The in-game crewmate slots ("Clint"/"Ben" in the manifest) keep their existing names; Clint's lines now clone Eads/Silver, Ben's lines now clone Smith/Hands. The reference clips are kept in the sound lab, not in this repository (pirate-sounds/ref/eads_silver_fixed.wav, pirate-sounds/ref/smith_hands_fixed.wav). Whisper reads all 42 lines back correctly within WER <= 0.15, except sails_2 (0.20) and anchor_2 (0.20) — both a single homophone slip ("sail(s)"/"anchor's"), not a real mispronunciation. 115 of the first 126 requested takes (42 lines x 3 seeds) came out; a small top-up run filled the 2 lines (damage_1, steering_2) that got zero usable takes in the first pass. Picks and WER scores: pirate-sounds/crew/eads_smith/picks.json.
A different, weaker real-voice source was also explored in parallel during this work and not used: Adrian Praetzellis's single-narrator LibriVox reading of Treasure Island, voicing both Billy Bones and Long John Silver himself (also Public Domain) — kept in the lab (pirate-sounds/crew_real/ in ~/Games/sound-lab/drama3/, picks in pirate-sounds/crew/real/picks.json) for the record, but one person reading two characters is a weaker match to "two distinct reference voices" than two different volunteers each performing a different character, so the Eads/Smith set above was installed instead.
Whether either take actually sounds convincingly like a real pirate to a human ear is unverified (nobody has listened), same caveat as every other voice decision in this file.