@audio/voice

Voice synthesis (speech analysis/processing lives in @audio/speech-*).

Package	What
@audio/voice-glottis	LF (Fant–Liljencrants–Lin) / Rosenberg glottal source — generator
@audio/voice-tract	Kelly-Lochbaum waveguide tract — Pink Trombone class (zakaton fork) — filter
@audio/voice-voder	Dudley 1939 Voder recreation (gmoe/voder) — frame sequencer

Three distinct signatures, chained glottis → tract for articulatory synthesis, or voder standalone:

import { glottis, tract, voder, VOWELS, BANDS } from '@audio/voice'

let excitation = glottis({ f0: 120, duration: 0.5 })       // → Float32Array (generator)
let vowel = tract(excitation, { shape: 'a' })               // Float32Array → Float32Array (filter)

let speech = voder([
	{ gains: [0,0,1,0,0,0,0,0,0,0], f0: 110, duration: 0.4 },  // frame: 10-band gains
])                                                            // frames[] → Float32Array

glottis(opts: {f0=120, duration=1, fs=44100, model='lf', Rd=1.7, jitter=0.005, shimmer=0.03, aspiration=0.01, seed}) → Float32Array

tract(excitation: Float32Array, opts: {shape='a' (vowel name in VOWELS | area array), sections=24, lipReflection=-0.85, glottalReflection=0.75, damping=0.996, fs=44100}) → Float32Array

voder(frames: Array<{gains: number[10], voiced=true, f0=110, duration=0.15}>, opts: {fs=44100, amp=0.8, seed}) → Float32Array — BANDS = the 10 channel center frequencies (225 Hz–7.5 kHz).

The site-todo "Voice generator (via natural tract gen)" idea. TTS = external/ML lane (see @audio/neural). Consumers: mridangam-syllable experiments, De-Slop resynthesis, speech-world.