BreezeTTS 2 · Voice Studio
Mode
Describe a voice in plain language. No recording needed. Best for creating a new character, narrator, host, or product voice.
01 · Script

Write the performance

Everything stays in one flow. Build the voice, render it, then listen below.

Language
0 / 900 characters—
Insert a vocal cue
02 · Voice

Define who is speaking

Start from a recipe or write a precise voice description from scratch.

Voice starting points
03 · Render

Generation settings

The defaults are deliberate. Open the controls only when you need them.

1 6

Take B keeps the script and voice brief, then changes only the seed.
Voice designZero-shot cloningVoice direction Automatic Whisper transcriptEnglish + Chinese 24 kHz monoStreaming
Build one believable speaker

Age, register, pacing, emotional state, and recording context are more useful than a long adjective list.

Correct the transcript

Whisper gives you a fast first pass. Review names, punctuation, vocal cues, and every spoken word before cloning.

Compare seeds, not settings

Enable take B to hear another stochastic interpretation without changing the underlying voice brief.

Load example
Mode Language Text to speak Voice description Direction strength

Breeze TTS 2 is an open-weight bilingual speech model from BreezeBlue. Reference transcription uses openai/whisper-large-v3-turbo.

The Breeze inference source is Apache-2.0. Breeze model weights, derivative models, and self-hosted outputs are licensed for research and non-commercial use only. Whisper is MIT licensed. Commercial Breeze use requires written authorization from RESONIA, INC. Only upload voices you have the right and consent to use. Do not use generated speech for impersonation, deception, fraud, harassment, or authentication bypass.