Phonon TTS ghcr.io/gradium-ai/ptts-openai-server:latest https://github.com/gradium-ai/xn-ptts/pkgs/container/ptts-openai-server latest Moves with each upstream release 0.5.0 Pinned to the 0.5.0 release bridge sh false https://github.com/Aurial/phonon-unraid/issues https://github.com/gradium-ai/xn-ptts https://raw.githubusercontent.com/Aurial/phonon-unraid/main/README.md Phonon TTS runs Gradium's Phonon text-to-speech models and serves an OpenAI-compatible speech API (POST /v1/audio/speech), so any client with an "OpenAI TTS" setting and a custom base URL works unchanged. It can also serve the Wyoming protocol, which Home Assistant uses for local voice. It runs entirely on the CPU and is fast: Gradium's models speak faster than real time even on a Raspberry Pi. Language: PTTS_LANG is required and picks how numbers and symbols are spelled out, and which catalog voices are served; it must match the language of the checkpoint (en by default; fr for a French model, and so on). First start: downloads the default model and the voices for its language into the cache folder, and logs little until they load; watch the log. The cache folder keeps them across restarts and updates, so later starts go straight to serving. Home Assistant: set PTTS_WYOMING_ADDR to 0.0.0.0:10200 and map the Wyoming port too; then add the Wyoming Protocol integration in Home Assistant with this server's address and port 10200. Access: the API has NO authentication on either port. Use it on a network you trust; do not expose it to the internet. Other models: leave Model empty for the built-in default, or name a Hugging Face repo as hf://owner/name (add @commit for a fixed version). The Phonon 40M models are gated: accept their terms on Hugging Face first and set an HF token here. Template and Unraid problems: the Support link. Phonon and the server itself: the Project link. AI:Tools:Utilities tts text to speech voice audio openai wyoming home assistant local voice assistant gradium rdna cpu https://raw.githubusercontent.com/Aurial/phonon-unraid/main/templates/phonon-tts.xml https://raw.githubusercontent.com/Aurial/phonon-unraid/main/icon.png Apache-2.0 **No authentication** — the API ports are open to anything that can reach them; run on a trusted network only. The first start downloads the model and voices (well under 1 GB) into the cache folder, which must persist. CPU only: no GPU or driver plugin needed; quantized weights (q8) do not use a GPU even when one is present. 8880 /mnt/user/appdata/phonon-tts en true q8