Say: echo but with TTS

Say uses coqui-TTS to create convincing voices for TTS application.

Flexible as you like it.

Installation

pip install tts-say
# Or from source
pip install git+https://gitlab.com/waser-technologies/technologies/say.git

Usage

From super simple...

❯ say Hello World
Hello World

...to choosing your own vocoder.

❯ say --help
usage: say [-h] [-n] [-e] [-E] [-v] [-L LANG] [--out_path OUT_PATH] [--list_models [LIST_MODELS]] [--model_name MODEL_NAME] [--vocoder_name VOCODER_NAME] [--config_path CONFIG_PATH] [--model_path MODEL_PATH] [--vocoder_path VOCODER_PATH] [--vocoder_config_path VOCODER_CONFIG_PATH] [--speaker_idx SPEAKER_IDX]
           [--speaker_wav SPEAKER_WAV [SPEAKER_WAV ...]] [--speakers_file_path SPEAKERS_FILE_PATH] [--use_cuda USE_CUDA] [--debug DEBUG]
           [text ...]

Same as echo but with Text-To-Speech.

positional arguments:
  text                  Text to be said.

options:
  -h, --help            show this help message and exit
  -n, --n               do not output the trailing newline
  -e, --e               enable interpretation of backslash escapes
  -E, --E               disable interpretation of backslash escapes (default)
  -v, --version         output version information and exit
  -L LANG, --lang LANG  Language to be spoken (default: $LANG)
  --out_path OUT_PATH   Output wav file path.
  --list_models [LIST_MODELS]
                        list available pre-trained tts and vocoder models.
  --model_name MODEL_NAME
                        Name of one of the pre-trained tts models in format <language>/<dataset>/<model_name>
  --vocoder_name VOCODER_NAME
                        name of one of the released vocoder models.
  --config_path CONFIG_PATH
                        Path to model config file.
  --model_path MODEL_PATH
                        Path to model file.
  --vocoder_path VOCODER_PATH
                        Path to vocoder model file. If it is not defined, model uses GL as vocoder. Please make sure that you installed vocoder library before (WaveRNN).
  --vocoder_config_path VOCODER_CONFIG_PATH
                        Path to vocoder model config file.
  --speaker_idx SPEAKER_IDX
                        Target speaker ID for a multi-speaker TTS model.
  --speaker_wav SPEAKER_WAV [SPEAKER_WAV ...]
                        wav file(s) to condition a multi-speaker TTS model with a Speaker Encoder. You can give multiple file paths. The d_vectors is computed as their average.
  --speakers_file_path SPEAKERS_FILE_PATH
                        JSON file for multi-speaker model.
  --use_cuda USE_CUDA   true to use CUDA.
  --debug DEBUG         true to enable debug mode.

say gives you the power.

Start the server

First you need to load the models in memory.

To do so, start the TTS server using say without any text argument.

say [--list_models [LIST_MODELS]] [--model_name MODEL_NAME] [--vocoder_name VOCODER_NAME] [--config_path CONFIG_PATH] [--model_path MODEL_PATH] [--vocoder_path VOCODER_PATH] [--vocoder_config_path VOCODER_CONFIG_PATH] [--speakers_file_path SPEAKERS_FILE_PATH] [--use_cuda USE_CUDA] [--debug DEBUG]
No attribute `text`.
say --help
For more information.
Starting server now.
Please wait.
...

Or enable its service.

cp ./speak.service.example /usr/lib/systemd/user/speak.service
systemctl --user enable --now speak.service

Get authorization to speak

You need to authorize the system to speak first. Change the service configuration as follows.

# ~/.assistant/tts.toml
...
[tts]
is_allowed = true
...

Then start the server and use say with some text argument to say something.

Use the client

Before you use the client, make sure :

1. the system has a valid authorization to speak,
2. the server has correctly loaded the models,
3. if the server has loaded YourTTS (by default); you need to create a style_wav file of your default speaker.

say [-n] [-e] [-E] [-v] [-L LANG] [--out_path OUT_PATH] [text ...]

❯ say --version
Say, version two dot, zero dot, three.
Say: version 2.0.3
Copyright (c) 2022, Danny Waser
TTS version 0.6.2
...

❯ say Hello, this is a test
Hello, this is a test

Save the audio

To save the resulted speech, use the argument --out_path.

❯ say "Bonjour." --out_path "say_output.wav"
Bonjour.
❯ soxi say_output.wav

Input File     : 'say_output.wav'
Channels       : 1
Sample Rate    : 16000
Precision      : 16-bit
Duration       : 00:00:01.17 = 18726 samples ~ 87.7781 CDDA sectors
File Size      : 37.5k
Bit Rate       : 256k
Sample Encoding: 16-bit Signed Integer PCM

Setup your own voice (YourTTS only)

By default, the server uses YourTTS to produce speech.

Therefor, before saying anything, you need to add a wav file to ~/.assistant/data/${lang}/TTS/styles/default.wav.

Where $lang is your target language (i.e en, fr, etc.).

This wav file must contain between 5 and 15 seconds of speech.

Make sure it matches with your tts.toml configuration.

You can also use the flag --speaker_wav manually.

say "Hello." --speaker_wav "~/.assistant/data/en/TTS/styles/default.wav"

Don't want to hunt down a voice?

Checkout my collection of high quality TTS voices generated using TTS VTCK/VITS models.

Audio samples

Yes yes but echo is for text right ?

Yes but you should be able to alias echo to say inside your favorite shell.

Because when you think about it, asking your computer to say something is like asking it to echo something.

Both cases output something.

Where echo repeat what it got in stdin, say as an injonction is used to ask someone to repeat what comes after.

Like so :

❯ Assistant, say Hello.
[Assistant] Hello.