Docker-TTS-API

AI text-to-speech and XTTS-2 Text-Based Voice Cloning

Docker container for xttsV2 with API, UI & voice cloning. This project is based on Coqui-TTS to make AI text to speech and voice cloning simple to install using docker. Seeing that most AI projects require python sknowledge to deploy, i decided to write use NodeJS/Javascript to create a simple to use AI Text to speech API. If you clone good quality voices, you'll get eleven labs quality. I found out that using 10 minutes of continous speech as a voice to clone gave me amazing Eleven labs quality results. You can clone with as short as 60 seconds audio but i found 10 minutes as the sweet spot that gave me the best results. Use podcast.adobe.com/enhance to clean any audio you want to clone.

Installation

After installation, the TTS engine may take some time to start up as it needs to download the models. This is one-time only.

git clone https://github.com/lojik-ng/docker-tts-api-ui.git
cd docker-tts-api-ui
mv server/keys.sample.json server/keys.json
docker build -t docker-tts-api-ui .
docker run -d -it -p 2902:2902 --gpus all  --restart=unless-stopped -v .:/shared -v ./models:/root/.local/share/tts --name docker-tts-api-ui docker-tts-api-ui

You can now access the ui at http://localhost:2902/.

Endpoints/API

• To get list of available default voice models, Send a GET request to http://localhost:2902/list-models
• To get list of available clone voices, Send a GET request to /list-voices
• To use a default voice model, send a POST request to http://localhost:2902/use-model with {prompt: string, apiKey: string, speaker: string, forceDownload: boolean}
• To generate with a cloned voice, send a POST request to http://localhost:2902/use-voice with {prompt: string, apiKey: string, speaker: string, forceDownload: boolean}
• Check server/index.html for example usage of the endpoints
• forceDownload will make the server return a downloadable stream if true or json with filename property if false. (See server/index.html)

Voice Cloning

• To clone a voice, copy a .wav audio file of the voice into /voices folder in the cloned repository. The docker container will use it from there.
• The filename of the audio file (without the extension - .wav) becomes the name of the cloned voice automatically.
• The list endoint of clone voice will automatically list it among available cloned voices.

API Keys/Authentication

• Edit /server/keys.json in the cloned repository anytime to add or remove API keys.

Logging

• logs are rotated daily and can be found in /logs folder of the cloned repository.
• Logs are never purged. You'll need to manually purge the logs.
• Access logs, error logs etc are lumped together

Features

• Requires GPU. I tested with Nvidia GPU (Cuda). I dont have AMD gpu to test.
• Not intended for parallel generation. I didnt test with parallel generation.
• API is written in NodeJS so should be easy for JS devs to modify. Check server/index.js.
• Returns mp3 files. I bundled it with ffmpeg to convert the generated .wav file into mp3 before sending. If you prefer .wav or any other audio file type, modify server/index.js
• Authentication: You can add as many users and ban users just by editing the server/keys.json file.
• Logging: It logs all user requests, errors etc.

Credits

• This software uses libraries from the FFmpeg project under the LGPLv2.1
• This software uses Coqui TTS