v1.2.0

Streamed generation, guided voice wizard, seekable player

Voice cloning made easy

• New Voices tab with a step-by-step wizard: pick a curated reading script (EN/DE) or write your own — the transcript fills itself in automatically
• Live level meter and waveform while recording (visible before you start, so you know the mic works), mic picker, silence detection, color-zoned 5–15 s timeline
• Review step with waveform preview, duration verdict, and an instant "Test this voice" sample
• Voices are analyzed once and cached — the 654 MB encoder no longer runs on every generation, so cloned-voice generation starts much faster
• Rename, preview, test, and delete saved voices; drag-and-drop upload

Streaming generation

• Text limit raised 500 → 2000 characters; long texts are split into sentence chunks and audio starts playing as soon as the first chunk is ready
• Voice stays consistent across chunks (chunk 1's tokens anchor the rest)
• Graceful cancel keeps the audio already generated — no more full engine reload
• 🔒 Voice lock: keep a random designed voice you like for the next generations

Player & polish

• Click-to-seek waveform with playhead and elapsed/total time
• Byte-accurate first-run download screen with speed and time remaining
• Library: delete items, reuse text, voice + quality info per generation; history storage now bounded
• Styled dialogs and toasts replace browser popups
• Keyboard navigation and screen-reader support throughout

Fixes

• Donate button no longer throws a console error
• Download progress tracked shard count instead of bytes
• Encoder load failure no longer breaks the whole app (cloning degrades gracefully)
• Duration estimation for voices saved without a transcript
• History/voice database clear races that could break persistence