Simple TTS

A simple machine learning text-to-speech program powered by kokoro.

Warning

This is currently only tested working on Ubuntu-based distros due to the required packages.

Setup

Clone repo and go into the directory

$ git clone https://git.ayo.run/ayo/simple-tts
$ cd simple-tts

Create new environment. Here I use conda.

$ conda create -n tts

### for Intel XPU specific device usage:
$ conda create -n tts --clone llm-pt26

Important

For using Intel XPUs, you need to set up ipex-llm environment with pytorch 2.6. Also, see Intel XPU environmental variables" section below.

Activate the environment and install the dependencies

$ conda activate tts
$ python -m pip install -r requirements.txt

Required packages

The following are required packages aside from the python dependencies. espeak-ng is used by kokoro under the hood for english languages, and libvlc is used as the default audio player for the generated audio.

$ sudo apt update
$ sudo apt install vlc espeak-ng

Note

Installing vlc via flatpak or snap will not work, as the code need access to libvlc.

Intel XPU environmental variables

For XPUs, we need to set some environmental variables. I have added a env.sh script which will activate the conda environment tts and set the environmental variables.

$ . env.sh

Usage

Go into the directory and activate the environment:

$ cd simple-tts
$ conda activate tts

If using Intel XPUs, set the env variables

$ . env.sh

Running the program without arguments will use the demo text tongue-twister.txt with the default voice.

$ python tts.py # will use default arguments

You can pass a string as first argument:

$ python tts.py "Hello world!" # will be read by the default voice

To run the program with an input file, use flag --input_file.

$ python tts.py --input_file demo/tongue-twister.txt

# or shorter...
$ python tts.py -i demo/tongue-twister.txt

Voices

Optionally, you can indicate a voice you want to use with the --voice flag. See all voices available.

$ python tts.py --voice am_michael

# or shorter...
$ python tts.py -v am_michael

There are four shortcuts available to the best voices: pro, hot, asmr, brit (i.e., best trained voices), and pro is the default if no value is given

$ python tts.py "Hello there!" --voice pro # af_heart

$ python tts.py "Hello there!" --voice hot # af_bella

$ python tts.py "Hello there!" --voice asmr # af_nicole

$ python tts.py "Hello there!" --voice brit # bf_emma

Disable audio player

You can disable the built-in audio player with --skip_play if you choose to play the audio files generated with your preferred player.

$ python tts.py "Hello there!" --voice asmr --skip_play

# or shorter...
$ python tts.py "Hello there!" --voice asmr -s

Demo Outputs

Voice: pro (ah_heart)

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_heart-0.wav

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_heart-1.wav

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_heart-2.wav

Voice: asmr (ah_nicole)

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_nicole-0.wav

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_nicole-1.wav

https://git.ayo.run/ayo/simple-tts/src/branch/main/demo/tongue-twister-af_nicole-2.wav