Higgs voice lab

Try natural speech, Nepali and English, voice cloning, and expressive delivery on the RTX 5090. Audio plays as it arrives.

Checking server…

Try an exampleVoiceText to speak

Emotion

Style

Pace

Pitch

Expression

Insert at cursor

Clone a voice from a reference clip

Upload a clean 5–15 second recording of a voice you have permission to use. Add its exact transcript for better fidelity. Reference uploads are deleted after the request.

Reference audioReference transcript

Sampling controls

Temperature

Top K

Maximum audio tokens

Seed

Ready.

First audio received—

Total generation time—

Audio duration—

Real-time factor—

Text sent to the model

First audio is measured in this browser, including network time. Real-time factor = generation time ÷ audio duration; lower is faster. VITS has been benchmarked and is stopped.

Evaluation lab • Boson Higgs TTS 3 • Test lab