WavTTS¶

Development line: project:wavtts · thread wavtts-development
Last event: 2026-06-08 · 1 dated since 2026-06-08 · Researched: 2026-09-05 · confidence: high

What it is¶

WavTTS is an open-source zero-shot text-to-speech system that conditions on a short reference recording and its transcript. Abilities: direct raw-waveform generation; CLI and script inference; 16 kHz official checkpoint; training and evaluation code. Limit: the released weights are CC BY-NC 4.0, and the model is not served by a Hugging Face Inference Provider. Verdict: use it for local non-commercial voice-conditioned synthesis, not as a drop-in hosted production API.

Development line¶

• 2026-06-08 — WavTTS public project resources were recorded. On 2026-06-08, the WavTTS development line was recorded with links to its project website, source repository, model page, demo, and a ComfyUI integration. This provides a dated public reference point for locating the project and related tooling. The supplied evidence does not establish what release or code change, if any, accompanied the links.

What changed¶

2026-06-08 — WavTTS was presented with its project page, official code and 16 kHz checkpoint, a web demo, and a later community ComfyUI integration; the official release itself was dated 2026-06-03.

How to use this¶

As of 2026-06-08, practitioners should use the recorded WavTTS project, source, model, demo, and ComfyUI links as the dated starting point for evaluating or integrating the project; no specific release behavior is established by this line alone.

1. Clone the official repository, create a Python 3.10 Conda environment, install CUDA-compatible PyTorch, then install WavTTS in editable mode. — https://github.com/cwx-worst-one/WavTTS
2. Run wavtts_infer-cli with --model WavTTS, a reference WAV, that recording’s transcript, and the text to synthesize; the default configuration downloads the official checkpoint automatically. — https://huggingface.co/worstchan/WavTTS
3. For ComfyUI, install the community node through ComfyUI Manager or into custom_nodes, restart ComfyUI, then connect a reference-audio input, required reference transcript, model loader, and generator node. — https://github.com/Saganaki22/WavTTS-ComfyUI

Best practices¶

• Use the supplied Conda/Python 3.10 setup and a CUDA-supported PyTorch build; the upstream repository recommends Conda for dependency management. — https://github.com/cwx-worst-one/WavTTS
• Supply an accurate transcript with the reference recording. The ComfyUI integration treats reference_text as required and recommends a clean 5–12 second voice prompt. — https://github.com/Saganaki22/WavTTS-ComfyUI
• For the community ComfyUI integration, prefer FP32 for reliability or its mixed-BF16 preset for lower VRAM; do not use FP16 presets, which its maintainer reports can yield non-finite samples. — https://github.com/Saganaki22/WavTTS-ComfyUI

Superseded by this¶

• Nothing marked obsolete yet.

Still unknown¶

• The Hugging Face Space linked on 2026-06-08 returned an internal error during review, so its current availability and exact interface are unverified.
• The ComfyUI integration is a separate community repository, not an official WavTTS release.

Sources¶

Agent brief¶

• Subject: project:wavtts, thread wavtts-development, 1 dated events 2026-06-08 → 2026-06-08.
• Practical note: As of 2026-06-08, practitioners should use the recorded WavTTS project, source, model, demo, and ComfyUI links as the dated starting point for evaluating or integrating the project; no specific release behavior is established by this line alone.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.