Researchers present TontaubeV1, a text-to-speech model that preserves natural prosody while enabling streaming inference from a single consumer GPU. The system uses a hierarchical DualCodec representation at 12.5 Hz to separate semantic streams from acoustic refinements, allowing for low-latency generation.
- Text is tokenized per character and processed by Qwen3-derived transformers to predict utterance duration.
- Three progressively smaller transformers add acoustic refinements, totaling 2.9B parameters.
- Overlapping DualCodec reconstructions are mapped into the VibeVoice latent space for causal decoding.
- The model accepts up to one minute of reference audio and supports English, German, and multilingual inputs.
- On a single RTX 5090, the streaming path reaches approximately 200 ms to first audio with an end-to-end RTF of 0.08.
- TontaubeV1 matches ElevenLabs Flash v2.5 and outperforms Fish Audio S2 Pro, Gradium API, and Cartesia Sonic 3 on prosody.
The model weights are released on Hugging Face under the Tontaube Community Model License 1.0.