Tontaube V1.1 brings local speech to smaller hardware, with limits behind the open-weight label
Tontaube's V1.1 inference release was tagged November 12. This is a November 12 release, not a fresh launch today.
The 4-bit model card describes a 1.8 GB download that supports Linux desktop CPUs as well as NVIDIA GPUs. The developer reports roughly real-time synthesis on a 12-core, dual-channel DDR4 desktop, while warning that long streams can pause between chunks. This is a vendor measurement, not a Guth test.
The inference repository targets long-form speech, streaming and reference-based voice cloning. It provides controls over delivery and supports English and German as its primary languages. Native macOS and Windows are not supported; Windows users are directed to WSL2. The server and model weights are separate artifacts, with matching release versions required.
The license and benchmark caveats matter
The model license, effective August 26, 2026, is not an unrestricted open-source license. Its summary permits free research and noncommercial use subject to restrictions, and free commercial use below a group-revenue threshold of EUR 200,000. Offering a speech-generation service requires a commercial license regardless of revenue, apart from a small free demo. Check the actual terms before deployment.
The repository's published comparison evaluates the earlier TontaubeV1, not V1.1. It uses an automated judge and warns that voice-reference differences can favor Tontaube. It does not establish general superiority or validate all languages.
Guth analysis: The useful development is a practical local speech option, not a new universal audio champion. For a pilot, check license eligibility, latency on your own hardware and exact text fidelity before considering cost savings. Use reference voices only with permission.
Evidence limits: No independent benchmark corroboration found. Model-card publication/update dates are not exposed in the retrieved pages. GitHub tag gives November 12, 15:19 without an exposed timezone. Full-precision model card is additional background, not cited evidence: https://huggingface.co/TontaubeAI/TontaubeV1.1.
Sources and citations
Revision history
Published .