home-assistant.io/source/_integrations/tts.markdown

title, description, ha_category, ha_release, ha_codeowners, ha_domain, ha_quality_scale, ha_platforms, ha_integration_type, related

title	description	ha_category	ha_release	ha_codeowners	ha_domain	ha_quality_scale	ha_platforms	ha_integration_type	related
Text-to-speech (TTS)	Instructions on how to set up text-to-speech (TTS) with Home Assistant.	Media source Text-to-speech	0.35	@home-assistant/core	tts	internal	notify	entity	url title https://www.home-assistant.io/integrations/#text-to-speech List of integrations using the TTS integration url title https://www.nabucasa.com/config/tts/ TTS with Home Assistant Cloud url title https://www.home-assistant.io/integrations/google_translate/ Google Translate TTS url title https://www.home-assistant.io/integrations/microsoft/ Microsoft TTS url title https://www.home-assistant.io/voice_control/ Home Assistant Assist

Text-to-speech (TTS) enables Home Assistant to speak to you.

{% include integrations/building_block_integration.md %}

See all TTS integrations using this building block for ways to use it in your automations. If you are using the Home Assistant voice assistant, Assist, Assist is using TTS when replying to you. Another way to use TTS is by using TTS with Home Assistant Cloud.

The state of a text-to-speech entity

The state of a text-to-speech {% term entity %} is a timestamp showing the date and time when text-to-speech was last used.

Screenshot showing the state of a text-to-speech entity in the developer tools.

Actions

Action speak

Modern platforms will create entities under the tts domain, where each entity represents one text-to-speech service provider. These entities may be used as targets for the tts.speak action.

The tts.speak action supports message, language, cache, media_player_entity_id and options options. The text that should be spoken is set with message, and the media player that should output the sound is selected with media_player_entity_id. The language can be set with language, using the format required by the target entity platform (refer to specific platform documentation). See cache section for information on cache option. Additional settings can be specified with the options option, which include preferred audio settings (see preferred audio settings section for more info) and further settings of the target entity platform, e.g., voice, motion, speed, etc. (refer to specific platform documentation for any supported settings).

action: tts.speak
target:
  entity_id: tts.example
data:
  media_player_entity_id: media_player.kitchen
  message: "May the force be with you."

Action say (legacy)

The say action supports message, language, cache and options options. The text that should be spoken is set with message. The language can be set with language, using the format required by the platform (refer to specific platform documentation). See cache section for information on cache option. Additional settings can be specified with the options option, which include preferred audio settings (see preferred audio settings section for more info) and further settings of the target platform, e.g., voice, motion, speed, etc. (refer to specific platform documentation for any supported settings). Since release 0.92, action name can be defined in configuration service_name option.

Say to all media_player entities:

# Replace google_translate_say with <platform>_say when you use a different platform.
action: tts.google_translate_say
data:
  entity_id: all
  message: "May the force be with you."

Say to the media_player.floor entity:

action: tts.google_translate_say
data:
  entity_id: media_player.floor
  message: "May the force be with you."

Say to the media_player.floor entity in French:

action: tts.google_translate_say
data:
  entity_id: media_player.floor
  message: "Que la force soit avec toi."
  language: "fr"

With a template:

{% raw %}

action: tts.google_translate_say
data:
  message: "Temperature is {{states('sensor.temperature')}}."
  cache: false

{% endraw %}

Cache

The integration cache can be controlled with the cache option in the action to speak or say, setting it to True to enable it (default), or False to disable it. A long time cache will be located on the file system. The in-memory cache for fast responses to media players will be auto-cleaned after a short period.

Preferred audio settings

Each TTS platform produces audio samples in different formats, not always compatible with every media player. TTS integration building block supports a way to configure preferred target audio format through options option of speak or say actions.

TTS integration building block uses FFmpeg integration to perform audio transcoding when target entity platform does not support one or all the specified preferred audio format settings (refer to specific platform documentation for any supported setting with related supported values).

Available preferred audio settings, all optional, are:

• preferred_format: Set the audio format. When not supported by the target entity platform, the value is a file extension like wav, mp3, ogg, etc., among ones supported by FFmpeg tool for output files.
• preferred_sample_rate: Set the sample rate. When not supported by the target entity platform, the value is in Hz as a number, among ones supported by the -ar parameter of FFmpeg tool.
• preferred_sample_channels: Set the number of audio channels. When not supported by the target entity platform, the value is a number among ones supported by the -ac parameter of FFmpeg tool.
• preferred_sample_bytes: Set the audio bit sampling. When not supported by the target entity platform, can only be set to 2 to use 16-bit audio sampling (any other value is ignored).

Example to produce an MP3 audio at 22050Hz:

action: tts.speak
target:
  entity_id: tts.example
data:
  media_player_entity_id: media_player.kitchen
  message: "May the force be with you."
  options:
    preferred_format: mp3
    preferred_sample_rate: 22050

REST API

POST /api/tts_get_url

Returns a URL to the generated TTS file. The engine_id (which is the entity id) or platform parameter together with message are required. Additional parameters cache, language and options are supported, as JSON attributes, as described for speak action.

{
  "engine_id": "tts.amazon_polly",
  "message": "I am speaking now"
}

The return code is 200 if the file is generated. The message body will contain a JSON object with the URL.

{
  "path": "/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_tts.demo.mp3",
  "url": "http://127.0.0.1:8123/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_tts.demo.mp3"
}

Sample curl command:

$ curl -X POST -H "Authorization: Bearer <ACCESS TOKEN>" \
       -H "Content-Type: application/json" \
       -d '{"message": "I am speaking now", "engine_id": "amazon_polly"}' \
       http://localhost:8123/api/tts_get_url

Troubleshooting

{% important %} Playing TTS media will prioritize the local Home Assistant URL, which can be configured by navigating to {% my network title="Settings > System > Network" %}. It is highly recommended to set the local Home Assistant URL to automatic, in which case the generated URL will be http://<local_ip>:<local_port>. {% endimportant %}

The following sections describe some of the problems encountered with media devices.

Self-signed certificates

This problem occurs when your Home Assistant instance is configured to be accessed through SSL, and you are using a self-signed certificate on your internal URL.

The tts action will send an https:// URL to the media device, which will check the certificate, and reject it. So it won't play your file. If you could make the device accept your certificate, it would play the file. However, many media devices do not allow changing settings to accept self-signed certificates. Ultimately, your option may be to serve files to local devices as http:// rather than https://.

Google cast devices

Google cast devices (Google Home, Chromecast, etc.) require the host in media URLs to be resolvable using Google's public DNS servers, and if the URL is specifying the https protocol, the certificate must be valid and not self-signed.

These requirements present the following problems, all of which create problems if the local Home Assistant URL is not http://<local_ip>:<local_port>:

• They reject self-signed certificates.

• They use Google's public DNS servers to resolve names, not DNS servers provided via DHCP. Additionally, they do not resolve local names through mDNS. This means they do not work with URLs that contain hostnames established by local naming means. Let's say your Home Assistant instance is running on a machine made known locally as ha (via local DNS) and homeassistant.local (via mDNS). All machines on your local network can access it as ha or homeassistant.local. However, try as you may, your cast device won't download the media files from your ha machine. That's because your cast device ignores your local naming setup. In this example, the say action creates a URL like http://ha/path/to/media.mp3 (or https://... if you are using SSL). If you are not using SSL, then setting an internal URL that contains the IP address of your server works around this issue. By using an IP address, the cast device does not have to resolve the hostname.

• If you are using SSL (e.g., https://yourhost.example.org/...) then you must use the hostname in the certificate (e.g., external_url: https://yourhost.example.org). You cannot use an IP address since the certificate won't be valid for the IP address, and the cast device will refuse the connection.

The recommended way to overcome these obstacles is to not manually configure a local Home Assistant URL.

Partial, corrupted or no audio

Some media players could reproduce only partial, corrupted or no audio at all when the audio format is not fully supported. In such cases it is required to experiment with different combinations of audio formats, channels, sample rates and bits using preferred audio settings options.

For example, some Google Cast devices skip initial audio part when the audio is sampled at 22050Hz, and to fix the problem it is required to set the preferred_sample_rate setting in the options option to 44100.