Silero TTS Service

Silero TTS Service

Content

• Information
• STRIVERTION OF THE server
• Server settings
• Setting in Home Assistant
• Settings in Rhasspy
• Functionality
• Endpoints
• Sound output on Bluetooth column

Information

I have created this project to provide my smart home with a normal speech synthesis. Also, to provide RHASSPY with a normal speech synthesis. Already ready -made solutions did not suit me and it was decided to invent their bike. The Silero models were taken as the basis.

I was inspired by the Silero-Ha-Http-TTS project from Gromina. He was damp and I decided to do everything in mind, with settings and ready -made containers.

STRIVERTION OF THE server

Installation via Docker:

Follow the command:

 docker run -p 9898:9898 -m 1g -e NUMBER_OF_THREADS=4 -e LANGUAGE=ru -e SAMPLE_RATE=48000 --name tts_silero -d navatusein/silero-tts-service

Installation via Docker Compose:

Create a docker-compose.yml file and transfer the contents to it:

 version : ' 3 '

services :
  silero-tts-service :
    image : " navatusein/silero-tts-service "
    container_name : " silero-tts-service "
    deploy :
      resources :
        limits :
          memory : 1G
    ports :
      - " 9898:9898 "
    restart : unless-stopped
    environment :
      NUMBER_OF_THREADS : 4
      LANGUAGE : ru
      SAMPLE_RATE : 48000

Follow the command:

 docker-compose up

Server settings

All server settings are transmitted as the Docker environment parameters to the container when starting.

The number of nuclei for speech processing NUMBER_OF_THREADS :

 NUMBER_OF_THREADS : 4 

The number of flows from 1 to the number of server processor cores.
By default: 4

Language synthesis LANGUAGE :

 LANGUAGE : ru

By default: ru

Supported languages ​​with votes available to them:

Language	Language code	Supported voices
Russian	ru	aidar baya kseniya xenia eugene random
Ukrainian	uk	mykyta random

SAMPLE_RATE sampling frequency:

 SAMPLE_RATE : 48000 

Possible values: 48000 , 24000 , 8000
By default: 48000

SOX SOX_PARAM Utility Parameters:

 SOX_PARAM : " reverb 50 50 10 " # Добавляет эхо на речь

By default: empty

The output file passes through the SOX utility. She can convey the parameters to impose effects on the speech: raise the timbre, add an echo, turn on the bass boost.

Link to the SOX utility documentation: https://linux.die.net/man/1/sox

Correction of the fraud of the end of the phrase HA_FIX :

 HA_FIX : True 

Can take values: True False
By default: False

Corrects a mistake in which Home Assistant does not agree on the end of the phrase. Adds a second of silence to the end of speech.

Setting in Home Assistant

In the configuration.yaml file, add a record:

 tts :
  - platform : marytts
    host : localhost # Адрес сервера
    port : 9898
    codec : WAVE_FILE
    voice : xenia # Имя голоса который хотите использовать.
    language : ru # Не используется. Настройки языка указываются в настройках сервера. 

Settings in Rhasspy Assistant

1. In the settings, in the Text to Spech section. Choose the Marrytts module.
2. Apply the RHASSPY Assistant settings.
3. Indicate the address of your server with the way /process .
4. Click on the Refresh button.
5. In the list of votes available, select the voice that you need.
6. Apply the RHASSPY Assistant settings.

Functionality

Normalization of numbers

The service can translate numbers into text.
Example:

 Текст с цифрой 1.

Normalization example 1

Declining nouns after the digit

The service can bow nouns after numbers.
To do this, the word that needs to be persuaded after the number, take <d>слово</d> tag.
Example:

 У меня было 15 <d>яблоко</d>.

Rlowing Example 1

If you need to persuade a few words, then each must be taken in the tag <d>слово</d> separately.

 Мне осталось работать 15 <d>рабочий</d> <d>день</d>.

Lange Example 2

Translite pronunciation

The service can pronounce translite.
Example:

 Lorem ipsum dolor sit amet.

Translite example 1

SSML

Using SSML, you can control the pauses and proxy synthesized speech.

 <p>
  Когда я просыпаюсь, <prosody rate="x-slow">я говорю довольно медленно</prosody>.
  Потом я начинаю говорить своим обычным голосом,
  <prosody pitch="x-high"> а могу говорить тоном выше </prosody>,
  или <prosody pitch="x-low">наоборот, ниже</prosody>.
  Потом, если повезет – <prosody rate="fast">я могу говорить и довольно быстро.</prosody>
  А еще я умею делать паузы любой длины, например две секунды <break time="2000ms"/>.
  <p>
    Также я умею делать паузы между параграфами.
  </p>
  <p>
    <s>И также я умею делать паузы между предложениями</s>
    <s>Вот например как сейчас</s>
  </p>
</p>

SSML Example 1

Endpoints

• GET /clear_cache - cleanses the cache of already synthesized messages.
• GET /settings - returns the current server settings.
• GET /voices - returns a list of available votes for the selected language.
• GET /process?VOICE=[Выбраный голос]&INPUT_TEXT=[Текст для обработки] - returns an audio file of synthesized speech.
• POST /process in the body of VOICE=[Выбраный голос] , INPUT_TEXT=[Текст для обработки] - returns an audio file of synthesized speech.

Sound output on Bluetooth column

1. If Home Assistant as the main OS (HAOS), then we read this TTS Bluetooth Speaker for Home Assistant documentation
2. If Home Assistant stands on Debian, then we do the following:

Edit Client.conf

 nano /etc/pulse/client.conf

Add the following:

 default-server = unix:/usr/share/hassio/audio/external/pulse.sock
autospawn = no

Restart Pulseaudio.

 pulseaudio -k && pulseaudio --start

We put an addid version of Current Version: 2.1.1 and put only this version. Mopidy 2.2.0 do not put - it is broken. Read more about the broken version of Mopidy 2.2.0 Read here.

Add to configuration.yaml

 media_player :
  - platform : mpd
    name : " MPD Mopidy "
    host : localhost
    port : 6600

We reboot Home Assistant completely to reboot Debian itself.

Connect the Bluetooth column to Debian, KB, J through Gui, or through the console using the Bluetoothctl command

Turn on Bluetooth:

 power on

Starting scanning devices:

 scan on

As we saw our device, we mate with the device:

 pair [mac адрес девайса]

We connect to the device:

 connect [mac адрес девайса]

Add the device to trusted:

 trust [mac адрес девайса]

Further, how Bluetooth devices added in two addons of Rhasspy Assistant and Mopidy you need to specify the source of the sound of the Bluetooth device:

1. In Rhasspy Assistant, we indicate this:

2. In Mopidy, we indicate this:

We check the performance:

Code:

 service : tts.marytts_say
data :
  entity_id : media_player.mpd_mopidy
  message : >-
    Спустя 15 лет жизнь некогда бороздившего космические просторы Жана-Люка
    Пикара