assisky

assisky

A Discord Speech-to-Text module made with Vosk. Yes, it works!

IMPORTANT!

In Discord settings, if you are not in the Push-to-Talk mode in "Voice & Video" and you are in the Voice Activity mode, you must disable "Automatically determine input sensivity" option and set it to lowest value. Like in this screenshot:

Installation

• Download a model. You can click here to find models.
• Extract the zip file to a folder.
• Rename the folder as "model" and put it in the root of your project or a path you want.
• If you put the model folder at a custom path then edit the model path in the code.
• Install ffmpeg. (apt install ffmpeg for Ubuntu)
• Install this npm module with npm i assisky

API

Assisky.setup(options)

options:

{
    "voskLogLevel": int, // -1 to disable logging of Vosk
    "modelPath": "Path to downloaded model folder", // Default: ./model
}

Returns an event for sending recognition results. See example for details.

Assisky.startListeningUser(userId, discordVoiceConnection)

userId: A Discord user ID that joined to a voice channel.

discordVoiceConnection: The voice connection of the bot. (Bot should join the same VC with the user.) See this for more details.

Assisky.stopListeningUser(userId)

userId: A Discord user ID that joined to a voice channel and recognition for user has been started and progressing.

Assisky.stopListeningChannel(channelId)

userId: A Discord channel ID, bot will stop listening everyone in the channel. Useful to run before leaving a VC.

Assisky.listeningList

Bot returns an Object which contains current listening users' streams, connection, userId. {discordAudio,PCMToMP3,wavReader,rec,connection,userId}

Assisky.config

The config object that recieved from Assisky.setup() function.

Example

Inspect the example folder.

Note: In the example below, I've used the Turkish language model but you can use different models as you want.
"bu bir denemedir" means "this is a test"
"merhaba dünya" means "hello world"