Install loudness-tts

1Check what you need

For Python: Python 3.9 or newer. For JavaScript: Node.js 20 or newer, Bun or Deno. Each voice is a one-time download of 20 to 140 MB; the default English voice is about 80 MB. FFmpeg is optional: it adds MP3, Opus, AAC and FLAC output, which chat apps such as LibreChat and AnythingLLM expect.

python3 --version
node --version

2Install the Python package

On macOS and Linux:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install loudness-tts

On Windows, in PowerShell:

py -m venv .venv
.venv\Scripts\activate
python -m pip install loudness-tts

For word timings taken from the voice model itself, which give the most exact subtitles, add the optional extra: python -m pip install "loudness-tts[timings]".

3Make your first speech

loudness-tts "Dr. Smith paid $5.50 on 4/5/2026 at 9:30 AM." -o speech.wav --srt speech.srt

The first run downloads the best English (US) voice and checks it against its published checksum; later runs work offline. Voices are kept in ~/.cache/loudness-tts (on Windows, %LOCALAPPDATA%\loudness-tts). Choose another language with --language es-ES, or a voice with --voice; loudness-tts voices lists them, best first.

From Python:

import loudness_tts

speech = loudness_tts.synthesize("The meeting starts at 9:30 AM.")
speech.save("meeting.wav")
speech.save_srt("meeting.srt")
print(speech.voice.credit)   # the credit line the voice licence asks for

4Or install the JavaScript package

Install the package file from the latest GitHub release. With Bun, use bun add with the same address.

npm install https://github.com/Loudness-ai/loudness-tts/releases/download/v0.1.1/loudness-tts-0.1.1.tgz

import { speak, synthesize } from "loudness-tts";

await speak("Hello! The meeting starts at 9:30 AM.", "hello.wav");
const speech = await synthesize("Hola, ¿qué tal?", { language: "es-ES" });
await speech.saveSrt("hola.srt");

In a web page, import it with your bundler (Vite, webpack or esbuild): speak(text) plays the speech and speech.blob() gives a WAV file. The browser keeps each voice after its first download.

5Start the speech server

loudness-tts serve --port 8880

The server answers like OpenAI’s speech API at http://127.0.0.1:8880/v1. Test it on macOS or Linux:

curl http://127.0.0.1:8880/v1/audio/speech -H "Content-Type: application/json" \
  -d '{"model": "tts-1", "input": "Hello from my own server!", "voice": "nova", "response_format": "wav"}' -o hello.wav

To let other computers or Docker containers use it, add --host 0.0.0.0 and protect it with --api-key and a secret of your choice. The app guides below show the settings for Open WebUI, LibreChat and AnythingLLM.

6Or run the server with Docker

docker run -d --name loudness-tts -p 8880:8880 -v loudness-tts:/data ghcr.io/loudness-ai/loudness-tts

The image includes FFmpeg and stores voices in the loudness-tts volume. To download languages when the container starts, add -e LOUDNESS_TTS_PRELOAD="en-US es-ES".

7Use it offline

Download the voices you need while you are online:

loudness-tts download --language en-US
loudness-tts download --language de-DE

Then set LOUDNESS_TTS_OFFLINE=1 to forbid downloads; LOUDNESS_TTS_HOME moves the voice folder. loudness-tts doctor --synthesize checks the whole installation.