Skip to content

Latest commit

 

History

History
101 lines (79 loc) · 8.59 KB

using_custom_voices_in_home_assistant_os.md

File metadata and controls

101 lines (79 loc) · 8.59 KB

How to install and use a custom Piper voice model in Home Assistant OS

I put together this guide because I couldn't find an adequate explanation of how to do this anywhere else. Please read carefully as there are currently some issues arising from the way Piper is implemented in Home Assistant that make this process more confusing than it should be.

Note: this guide assumes you will run your voices on the same physical device (eg. Raspberry pi) as Home Assistant.

  • This option is best for people who plan to render short phrases or phrases that will be used repeatedly.
  • To learn how to run a (GPU accelerated) piper docker container on another machine on your network, use this guide instead.

Install Piper add-on and Wyoming protocol integration

  1. Install the Piper Add-on (Settings > Addons > click ADD-ON STORE button > search for Piper.)
  2. Install the Wyoming protocol integration (Settings > Devices & Services > Integrations > Click ADD INTEGRATION BUTTON > search for Wyoming Protocol).
  3. When prompted by Wyoming Protocol for host and port, you can use core-piper for host and 10200 for port
    • core-piper is the hostname provided by the piper add-on's docker container. You can also use the ip address of the machine running the docker container.
    • 10200 is the default port for Piper.

Prepare your custom voice models for upload

Note: The latest version of TextyMcSpeechy already exports voices that are ready to use.

  • If your voice files are named something like en_US-bob_1234-medium.onnx, you can use them as-is.
  • Voices created with earlier versions of TextyMcSpeechy did not comply with Piper's expected naming convention. In order for your voices to be detected and appear in menus properly, you will need to rename them using the system prescribed by Piper. You may also need to edit the .onnx.json file.
  • Instructions for doing this can be found here.

Upload your .onnx and .onnx.json files to /share/piper

  • This is a bit challenging on Home Assistant OS since you don't have permissions to upload to this folder via the webui.
  • There are a variety of ways of accomplishing this which are beyond the scope of this guide.
  • If you don't find a piper directory inside /share, create it yourself.
  • I did this by installing the FTP Add-on (Settings > Add-ons > click ADD-ON STORE button > search for FTP)
  • After setting up credentials in the FTP add-on and ensuring it was up and running, I used Filezilla (FTP client software) on my PC to connect to the server and upload my model files to /share/piper.

How do I use my custom models?

Once your .onnx and .onnx.json files are in the /share/piper folder, restart the Piper add-on and reload the Wyoming Protocol integration, otherwise they won't know about your models.

  • If you have done everything correctly so far, your voice will technically be ready to use, but there are some implementation issues you need to know about.
  • Important: Due to the way the Piper add-on currently gets its voice lists, if you go to Settings > Add-ons > Piper > Configuration and look for your custom voice in the dropdown menu, you won't find your custom voices there. I have opened an issue for this on github.
  • The only place you are currently able to see your custom voice in the webui is in Settings > Voice Assistants, after creating or modifying a voice assistant entity using Piper as the text to speech engine.
  • There is a "Try Voice" button you can use to verify that your custom voice is working properly. I recommend doing this before proceeding further.
  • If your custom model shows up with the wrong name, you probably need to change some fields in your .onnx.json file and restart Piper/Wyoming protocol.

How do I use my custom voices in my own scripts?

This guide explains how to create and call a service script that will render text using your custom voices. You should create one for each of the voices you intend to use.

A simple TTS voice tester GUI for Home Assistant lovelace dashboards

  1. Create a dropdown list to contain the names of all your custom voice services

    • In Settings > Devices & Services > Helpers tab, click CREATE HELPER button and choose Dropdown
    • In the name field, enter tts_voices (this tutorial will assume that the dropdown's entity name will be input_select.tts_voices)
    • in the options field, add the name of the first voice you want to be able to choose from, prefaced by script., eg. script.say_as_bob
    • add the rest of the voices and save the dropdown list. You can add more voices to this list later.
    • important: immediately after your dropdown is created, there might not be any voice selected, and this would cause the script to fail. Click the tts_voices entity you just created and in the popup window, choose one of your voices from the dropdown list.

  2. Create a text input box to hold the demo text you want the TTS engine to say.

    • In Settings > Devices & Services > Helpers tab, click CREATE HELPER button and choose Text
    • In the name field, enter text_to_say (this tutorial will assume that the text entity's name will be input_text.text_to_say)
    • click on the newly created entity and in the popup window, enter a test phrase in the text input box and save it.

  3. Create a button to say your test phrase in the selected voice. This will be configured later.

    • In Settings > Devices & Services > Helpers tab, click CREATE HELPER button and choose Button
    • In the name field, enter say_it (this tutorial will assume that the button entity's name will be input_button.say_it)
    • Click Create

  4. Create a script that will say the text in your text box using the voice service selected in the dropdown list.

    • In Settings > Automations and Scenes > Scripts tab, click the CREATE SCRIPT button, choose Create new script.
    • From the kebab menu (three vertical dots in the top right corner), choose Edit in YAML
    • Delete the line that says sequence: []
    • Paste in the following script:
sequence:
  - data_template:
      message: "{{ states('input_text.text_to_say') }}"
    action: "{{ states('input_select.tts_voices') }}"
alias: Test TTS
description: Say text in the input box
  • Click the SAVE SCRIPT button. Name this script test_tts (this tutorial will assume that this script entity's name will be scripts.test_tts)
  • Once your test_tts script has been created, you should be able to test it. Click the kebab menu (three vertical dots) on the line associated with the test_tts script, and choose Run. If you have done everything right you should hear your test phrase spoken in the voice you have chosen.
  1. Create entities in your lovelace dashboard and configure the button image
  • In the example above, an Entities card was created by putting the dashboard into edit mode, clicking the + button, and choosing the Entities card from the BY CARD tab.
  • in the popup window, under Entities (required), delete the example entities, then:
    • Set the first entity to input_select.tts_voices
    • Set the second entity to input_text.text_to_say
    • Click Save
  • To add the button, click the + button below the dropdown and text box you just added and in the BY CARD tab, search for the Button card and click it.
  • To configure the button, first replace anything in the Entity field with the button entity you created earlier (input_button.say_it)
  • Optionally give your button a name and/or icon.
  • In the Tap behavior field, choose Perform action
  • A new Action field will appear. Enter script.test_tts here to cause this button to trigger the script you created earlier.
  • Change any additional layout options as you wish. The example above has Full width card turned on in the Layout tab
  • click SAVE to save your button.
  • click the DONE button to stop editing your dashboard.
  • You now can say anything in any voice via your home assistant dashboard. Enjoy!

Known issues:

  • Sometimes the Say it button will play the previous message that had been stored in the text box rather than the one you have just typed. This is because the text box doesn't pass its new value to the rest of the system until it loses focus. Until I find a way to resolve this, you can click or tap anywhere outside the box after entering a new message before you click the button.