More reliable voice model loading. Loudness measured to the international standard
Voice generation models now load reliably, and loudness normalization follows the international measurement standard so your export lands on the target you choose. Long audio is processed without freezing the page.
NewVoice generation WAV / MP3 exports now record that the audio is AI-generated (kept when you finish it in the audio editor)
NewStop a running transcription with "Cancel"
NewTrim handles, switches, and adding files work with a keyboard and screen readers
ImprovedLoudness is now measured per the international standard (ITU-R BS.1770), so exports match the target (LUFS) you choose, with clipping prevented after MP3 conversion
ImprovedThe page no longer freezes while adjusting loudness on long audio, and memory no longer keeps growing when you process repeatedly
ImprovedAI model downloads resume where they left off if interrupted, and you are told right away when storage is full
ImprovedPages load faster
FixedFixed voice model loading getting stuck at "Compiling"
FixedFixed transcription failing on devices where the GPU can't be used
FixedFixed transcription results and cuts applying to a different file or range
FixedFixed generation continuing with the previous voice when replacing the reference audio failed