ischeinkman / assistant-rs

AssistantRS

AssistantRS is a simple, configurable, and fully offline voice control program powered by Mozilla DeepSpeech and Rust. It can either be run as a one-off to run a single command, or can be run in the background for continuous assistance without having to reload constantly.

Usage

The program accepts the following flags:

• --config <file path> -- Read configuration from <file path>. Can be passed more than once to read multiple config files. See Configuration for more information.
• --daemonize or -d -- Run in "daemon" mode.
• --help or -h -- Outputs usage information and exits.
• --version or -V -- Outputs version information and exits.

By default, the program loads the configuration and model, listens for audio until it no longer detects human speech, runs the command closest to the detected message, and exits. When the -d flag is passed, the program loads the config and then sleeps until it recieves a Unix signal before responding as follows:

• SIGCONT | SIGUSR1 -- The program wakes up, listens and spawns a single command (following the same process as the default standalone mode), and sleeps again.
• SIGHUP -- The program wakes up, re-reads the config files (including the ones previously passed via the --config flag), and reloads the model if necessary.

Configuration

Configuration is done via toml files. The following fields are defined:

Field Name	Type	Description	Required?	Default
library-path	String	The path to the libdeepspeech.so library file to use.	No	None; libdeepspeech.so is loaded via the system loader.
model-path	String	The path to the DeepSpeech model file to use. Note that this must be compatible with the library passed to libdeepspeech.so.	Yes	None
scorer-path	String	The path to the DeepSpeech external scorer file to use.	No	The default scorer built in to DeepSpeech.
beam-width	Integer	A field internal to Mozilla DeepSpeech that controls the speed-vs-accuracy amount. This should usually only be increased if the assistant is having trouble accurately matching spoken commands to the list of valid commands.	No	1
command	List of {message : String, command : String}	A list of Commands, each containing a message keyphrase that the assistant listens for and a command that the assistant runs as a sub-process if it hears the keyphrase.	Yes	None

An example is included in /res/config.toml.

AssistantRS follows the FreeDesktop XDG config spec; it will read configuration options from the following paths in order, if they exist:

1. All paths passed via the --configs command line flag.
2. If the environment variable XDG_CONFIG_HOME is defined, then $XDG_CONFIG_HOME/assistant-rs/assistant.toml; otherwise $HOME/.config/assistant-rs/assistant.toml is read.
3. If the environment variable XDG_CONFIG_DIRS is defined, then it is treated as a list of directories separated by :; for each directory $DIR in this list, the config file $DIR/assistant-rs/assistant.toml is read; otherwise /etc/xdg/assistant-rs/assistant.toml is read.