Reverb reads text in spoken form, so the card now says normalizing the text comes before
generating: the package does it with Verbalis by default, and the ExecuTorch programs and
ruNNtime need it done first. The ruNNtime link points at its text-to-speech guide.
QNN int4 backbones: memory-planned outputs (run without output binding; same graphs and results)
The quality table now shows the leaderboard's MiniMax results, WER with the default voice
and SIM-o with voice cloning, against well-known models on the leaderboard. The speed table
gains an NVIDIA H200 row, and the header and the size and speed chart are re-rendered.
Reverb-TTS v1.0.0
IS02+4: benito47, kamkacz, claude opus 5.5, IgorSwat
f37e30c
Co-authored-by: benito47 <benito47@users.noreply.huggingface.co>
Co-authored-by: kamkacz <kamkacz@users.noreply.huggingface.co>
Co-authored-by: claude opus 5.5 <claude opus 5.5@users.noreply.huggingface.co>
Co-authored-by: IgorSwat <IgorSwat@users.noreply.huggingface.co>