ttsprech

A simple text to speech tool for the command line using silero-models or coqui-ai/tts.

Output is either provided in the form of .wav files or directly as audio to the speakers. Generation is fast enough for real time use (e.g. narrating longer text), but it will take a couple of seconds before the first words are said, making it unsuitable for fast responding interactive use.

It runs locally on the CPU. Neither a GPU or any kind of cloud services are required. Language models, other than the default, will however be downloaded on demand when requested.

Required language model files are downloaded on demand and stored in ~/.cache/ttspeech/.

Installation

ttsprech is distributed as Nix flake and can be run directly from the git repository:

$ nix run github:Grumbel/ttsprech

Nix flakes are still an experimental feature, so a reasonably new version of Nix and some configuration might be required. See Nix Flakes, Part 1: An Introduction And Tutorial.

NixOS is not required, the Nix package manager can be installed on other Linux distributions.

TTS Engines

• silero-models provides high quality and reliable voice output, much better than the robotic sounding espeak. It is the default.

• coqui-ai/tts provides even better sounding output than silero, but is much less reliable and a bit slower. Overly complex sentence structure will cause it to fail. Glitches at the end of the audio output are common.

Usage

usage: ttsprech [-h] [-v] [-f FILE] [-e ENGINE] [--ssml] [-m FILE] [-l LANGUAGE] [-s SPEAKER]
                [-r RATE] [-S NUM] [-E NUM] [-T NUM] [-O DIR]
                [TEXT ...]

Text to Speech

positional arguments:
  TEXT

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         Be more verbose
  -f FILE, --file FILE  Convert content of FILE to wav
  -e ENGINE, --engine ENGINE
                        Select the TTS engine to use (coqui, silero)
  --ssml                Interpret text input as SSML
  -m FILE, --model FILE
                        Model file to use
  -l LANGUAGE, --lang LANGUAGE
                        Use language LANGUAGE (default: auto)
  -s SPEAKER, --speaker SPEAKER
                        Speaker to use
  -r RATE, --rate RATE  Sample rate
  -S NUM, --start NUM   Start at sentence NUM
  -E NUM, --end NUM     Stop at sentence NUM
  -T NUM, --threads NUM
                        Number of threads to use
  -O DIR, --output-dir DIR
                        Write .wav files to DIR

Legal

Note that even so this software is covered under the GPLv3+, the silero-models used are under Creative Common Attribution-NonCommercial-ShareAlike 4.0 International.