nacatamalon/speech makes a voice sample by sample, the way the machines of the eighties talked: a throat that buzzes at a pitch, five formants that shape it into vowels, and noise for the s, the stops and the breath. In 1961 a computer sang Daisy Bell this way, the first song a computer sang.
The song's voice is a vocal part, daisy-bell.vocal, loaded with useLoadVocal and sung over the band with accompany: useSong plays them together, made as they play, so nothing waits to load. The same part is sung by another voice with a different throat, and lyricsAt gives the words as they are sung.
renderSpeech says text, and hands back a sound for useSound. Spanish is read by rules, and textToPhonemes shows how; anything else is written as phonemes. 1 to 4 play a take, space stops it.