model.ts: rename ONNX input ref_s → style for kokoro-js / kokoro-onnx interop

ui: group voice picker by gender via <optgroup> (24 voices was unwieldy as flat list)

voices: ship Sama (समा) + Navya (नव्या) — synthetic voicepacks, no source audio

voices: ship 17 stock-Kokoro crossover voicepacks (15 adult + 2 kid)

voicepack: Svara spelled स्वरा (correct Marathi feminine form, was स्वर)

voicepacks: graduate Svara + Vihaan from experimental → stable

voicepack: af_heart Maitri → Svara (स्वर) — 'sva-ra', musical note / voice

voicepacks: rename Heart→Maitri (मैत्री), Omega→Vihaan (विहान)

voices: add stock Kokoro af_heart + hm_omega as experimental crossover voices

synthesize: strip BOS+EOS audio from output (keep BOS in model input)

synthesize: prepend BOS/EOS to inputIds (was missing — caused leading-consonant eat)

audio: prepend 80ms leading silence to fix SpringLab utterance-initial consonant eat

punctuation pauses: splice silence after , ; : — . ! ? …

voicepacks: mf_asha + mm_vivek default_speed -> 1.0 (Rasa voices, neutral pacing)

metrics: TTFB (time to first audio) replaces total-synth in headline metric

ui: performance metrics card + voice ref-clip download + WAV download

voicepack: re-extract mf_mukta only (Stage 2 only + untrimmed SpringLab)

Revert "voicepacks: re-extract mf_mukta + mm_dnyanesh from trimmed SpringLab refs"

voicepacks: re-extract mf_mukta + mm_dnyanesh from trimmed SpringLab refs

loanwords: add India-specific brands + services

loanwords: hand-fix mispronouncing entries + add 6 missing common words

voicepacks: mf_mukta + mm_dnyanesh default_speed 0.9 -> 0.8 (SpringLab paces faster than Rasa; +12% frames per phoneme to un-eat first-word consonants)

speed slider now scales pred_dur (was playback-rate only)

Revert "voicepacks: re-extract from v0.5 kokorobase ep4"

voicepacks: re-extract from v0.5 kokorobase ep4

feat: per-word display shows user's original Roman input for substituted loanwords

cleanup: remove verification marker

fix: actually deploy textarea-preserving build (build entry was contaminated by previous dist)

deploy: actual textarea-preserving bundle (vite cache had stale build)

src: align main.ts source with deploy (in-memory transliterate)

preprocessor: keep textarea showing user input; transliterate in-memory only

loanword_map: movie → मूव्ही (was मोव्ही, sounded as 'moovie')

fix: deploy actual preprocessor bundle (3rd time the charm)

fix: deploy actual preprocessor bundle (previous push had pre-rsync build)

loanword preprocessor + re-exported v0.2 ONNX (torch 2.6)

v0.2 polish: per-voice default speeds + trailing audio pad

v0.2 release: built artifacts + source

v0.2 final release: 4 voices, v0.2 model

voicepacks: gender-paired chips + dynamic textarea default

release: webgpu demo + source

use sanscript syncope:true instead of regex halant-strip

add Latin→Devanagari transliterate button (ITRANS via sanscript.js)

header link color matches footer (gerua + kumkum hover)

Asha+Vivek (.bin) re-extracted from trim_silence-cleaned Rasa

voicepacks (.bin) re-extracted from Stage 2 final

remove bottom-strip rotation — figures were upside down; SVG already has triangle borders on both edges

restore max-width 1600px cap on warli strip; keep 180° bottom rotation

warli: show full strip at natural aspect, no crop; bottom strip 180° rotated

warli: 24px top breathing room + 180° rotated bottom strip

vectorize the rich Warli strip (PNG → SVG via potrace, 7× smaller)