Carsten Graf 2d4831349b feat(tts): browser-based MP3 announcer for finished times ...

Plays the just-finished time via Web Audio API on index.html and on
new top-1 entries on leaderboard.html. Snippets are pre-rendered as
German neural-TTS MP3s (numbers 0-99 spoken naturally as
"vierzehn", "sechsundneunzig" etc.) and decoded into AudioBuffers
once at page load, then chained gaplessly via start(when, offset,
duration) — leading/trailing silence in each MP3 is detected and
skipped so words flow without pauses. A floating speaker toggle
persists in localStorage and doubles as the user gesture that
unlocks the AudioContext on autoplay-restricted browsers (SmartTV,
iOS Safari).

Hundredths formatting mirrors the ESP's float-truncation via
Math.fround so the announced value always matches the displayed
string, even at hundredths boundaries where double/float rounding
diverges. Preload runs at concurrency 2 with a 2 s start delay so
the 107 MP3 fetches don't starve /api/data and freeze the live
timer.

Regenerator script: tools/generate-tts.py (requires edge-tts).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-03 20:21:49 +02:00

8.2 KiB

Raw History

View Raw