Wyoming Microsoft STT

Wyoming protocol server for Microsoft Azure speech-to-text.

This Python package provides a Wyoming integration for Microsoft Azure speech-to-text and can be directly used with Home Assistant voice and Rhasspy.

Azure Speech Service

This program uses Microsoft Azure Speech Service. You can sign up to a free Azure account which comes with free tier of 5 audio hours per month, this should be enough for running a voice assistant as each command is relatively short. Once this amount is exceeded Azure could charge you for each second used (Current pricing is $0.36 per audio hour). I am not responsible for any incurred charges and recommend you set up a spending limit to reduce your exposure. However, for normal usage the free tier could suffice and the resource should not switch to a paid service automatically.

If you have not set up a speech resource, you can follow the instructions below. (you only need to do this once and works both for Speech-to-Text and Text-to-Speech)

1. Sign in or create an account on portal.azure.com.
2. Create a subscription by searching for subscription in the search bar. Consult Microsoft Learn for more information.
3. Create a speech resource by searching for speech service.
4. Select the subscription you created, pick or create a resource group, select a region, pick an identifiable name, and select the pricing tier (you probably want Free F0)
5. Once created, copy one of the keys from the speech service page. You will need this to run this program.

Usage

Depending on the installation method parameters are parsed differently. However, the same options are used for each of the installation methods and can be found in the table below. Your service region and subscription key can be found on the speech service resource page (step 5 the Azure Speech service instructions).

For the bare-metal Python install the program is run as follows:

python -m wyoming-microsoft-stt --<key> <value>

Key	Optional	Description
service-region	No	Azure service region e.g., uksouth
subscription-key	No	Azure subscription key
language	Yes	Default language to set for transcription, default: en-GB. For auto-detection provide multiple languages.
uri	No	Uri where the server will be broadcasted e.g., tcp://0.0.0.0:10300
download-dir	Yes	Directory to download models into (default: )
update-languages	Yes	Download latest languages.json during startup
debug	Yes	Log debug messages

Multi-language support

This add-on can also auto-detect the spoken language from a list of pre-defined languages (max. 10). To do this in Home Assistant provide the languages separated by semi-colons like so:

Note

Setting multiple languages will override the options set by Home Assistant's Voice configuration! It will prompt you to select a language but the option is ignored when speech is processed.

Installation

Depending on your use case there are different installation options.

• Using pip Clone the repository and install the package using pip. Please note the platform requirements as noted here.

  pip install .

• Home Assistant Add-On Add the following repository as an add-on repository to your Home Assistant, or click the button below. https://github.com/hugobloem/homeassistant-addons

  

• Docker container To run as a Docker container use the following command:

  docker run ghcr.io/hugobloem/wyoming-microsoft-stt-noha:latest --<key> <value>

  For the relevant keys please look at the table below