ZenTTS

By Omor

English text-to-speech from your terminal. ZenTTS turns .txt, .epub and .pdf files into .wav or .mp3 audio, with per-chapter splitting, voice blending and live streaming playback.

• English only — US (en-us) and UK (en-gb)
• 100% offline after the first run, no API keys, no accounts
• Self-contained engine: the ONNX runtime and the phonemizer are all ZenTTS needs

Install

pip install zentts

Or from a clone of this repository:

git clone https://github.com/OmorDeveloper/zentts.git
cd zentts
pip install -e .

Model files

ZenTTS needs two files:

They download automatically the first time you run ZenTTS, with a progress bar and a checksum check, and are cached for later runs. To fetch them ahead of time:

zentts --download

The cache lives in %LOCALAPPDATA%\zentts on Windows, ~/Library/Caches/zentts on macOS and ~/.cache/zentts on Linux.

You can always point at your own copies:

zentts input.txt out.wav --model ./zentts-v1.0.onnx --voices ./zentts-voices-v1.0.bin

Usage

zentts input.txt output.wav --speed 1.2 --lang en-us --voice zen_us_f10
zentts part1.txt part2.txt part3.txt book.wav
zentts intro.txt chapters.epub notes.pdf audiobook.mp3 --format mp3
zentts book.epub --split-output ./chunks/ --format mp3
zentts paper.pdf output.wav --lang en-gb --voice zen_uk_f02
zentts input.txt --stream --voice "zen_us_f10:60,zen_us_m01:40"
zentts --merge-chunks --split-output ./chunks/ --format wav
zentts --help-voices
zentts --help-languages
zentts --help

Options

Voices

There are 28 ZenTTS voices. A voice id reads zen_<region>_<gender><number>:

Run zentts --help-voices to list them grouped by region and gender.

Blend two voices by weight:

zentts input.txt out.wav --voice "zen_us_f10:60,zen_us_m01:40"
zentts input.txt out.wav --voice "zen_us_m01,zen_us_f10"   # 50-50

Run it as a studio and API server

zentts start gives you two things at once: a web studio in your browser and an OpenAI-compatible API. Everything runs on your own machine — no account, no network once the model is downloaded, and no extra dependencies.

zentts start                                   # http://127.0.0.1:8000
zentts start --host 0.0.0.0 --port 8080        # reachable from your phone
zentts start --api-key secret                  # require a bearer token

The web studio

Open http://127.0.0.1:8000 in any browser, on the same machine or from a phone on the same network when you bind 0.0.0.0. It gives you:

• a text box that takes any length of text
• every voice, grouped by accent and gender, with speed, language and format
• Generate for a finished file, Stream to start hearing it immediately
• Sessions in the sidebar: create, rename and delete
• History per session: replay or download anything you made before
• Logs: the recent requests the server handled

History is saved on your own machine, under <ZENTTS_HOME>/sessions.

Endpoints

Generating speech

zentts start serves an OpenAI-compatible speech API, so anything already written against OpenAI's text-to-speech works against your own machine with a one-line change. No extra dependencies — the server is part of the package.

zentts start                                   # http://127.0.0.1:8000
zentts start --host 0.0.0.0 --port 8080        # reachable on your network
zentts start --api-key secret                  # require a bearer token
zentts start --default-voice zen_uk_f02

Endpoints

Generating speech

curl http://127.0.0.1:8000/v1/audio/speech   -H "Content-Type: application/json"   -d '{"input": "Hello from ZenTTS.", "voice": "zen_us_f10"}'   --output hello.mp3

Long text needs no special handling: it is split at sentence boundaries, spoken in order, and joined into one file.

Sessions and history

# make a session and generate into it
SID=$(curl -s -X POST localhost:8000/v1/sessions \
  -H "Content-Type: application/json" -d '{"name":"My audiobook"}' | jq -r .id)

curl -s localhost:8000/v1/audio/speech -H "Content-Type: application/json" \
  -d "{\"input\": \"Chapter one.\", \"session_id\": \"$SID\"}" --output ch1.mp3

curl -s localhost:8000/v1/sessions/$SID                    # history
curl -s -X PATCH localhost:8000/v1/sessions/$SID \
  -H "Content-Type: application/json" -d '{"name":"Renamed"}'
curl -s -X DELETE localhost:8000/v1/sessions/$SID          # delete it and its audio

Streaming

curl -N localhost:8000/v1/audio/speech -H "Content-Type: application/json" \
  -d '{"input": "Audio arrives as it is made.", "stream": true}' --output live.mp3

Audio is sent in chunks as each part is synthesised, so playback can start before the whole text is done. Streaming supports mp3 and pcm.

Using an OpenAI client

Point the base URL at ZenTTS and nothing else changes:

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8000/v1", api_key="not-needed")

response = client.audio.speech.create(
    model="tts-1",          # any model id is accepted
    voice="nova",           # OpenAI names map onto ZenTTS voices
    input="Running locally, with no API bill.",
)
response.stream_to_file("hello.mp3")

OpenAI voice names are mapped to their closest ZenTTS voice: alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer and verse. You can also pass ZenTTS ids directly.

Server options

Run zentts start --help for the full list.

Multiple input files

Pass as many inputs as you like and they are joined, in order, into one output file. Formats can be mixed. If the last argument ends in .wav or .mp3 it is the output; otherwise the name is taken from the first input.

zentts ch1.txt ch2.txt ch3.txt book.wav
zentts intro.txt body.epub appendix.pdf audiobook.mp3 --format mp3
zentts ch1.txt ch2.txt --stream          # plays them back to back

Long books

--split-output writes chapter_001/chunk_001.wav, … and skips work that is already done, so an interrupted run resumes where it stopped. When it finishes, join everything into one file per chapter:

zentts --merge-chunks --split-output ./chunks/ --format wav

Use it from Python

from zentts import ZenTTS, resolve_model_files
import soundfile as sf

model, voices = resolve_model_files()   # downloads on first use
engine = ZenTTS(model, voices)

samples, sample_rate = engine.create("Hello from ZenTTS.", voice="zen_us_f10")
sf.write("hello.wav", samples, sample_rate)

Requirements

Python 3.11 or later. Everything else is installed by pip. --format mp3 needs libsndfile with MP3 support, which ships with the soundfile wheels on Windows, macOS and Linux.

Writing audio to a file works anywhere. --stream plays through your speakers, so it additionally needs the PortAudio system library — already present on Windows and macOS, and installed on Debian or Ubuntu with sudo apt install libportaudio2.

License

ZenTTS is proprietary software, © 2026 Omor, all rights reserved — see LICENSE. You may use it, including commercially, but not redistribute or modify it without permission.

Third-party components and the model weights it builds on keep their own licences, reproduced in NOTICE.md.

Author

Omor — LinkedIn · GitHub