ovos-core

OpenVoiceOS is an open source platform for smart speakers and other voice-centric devices.

OpenVoiceOS is fully modular. Furthermore, common components have been repackaged as plugins. That means it isn't just a great assistant on its own, but also a pretty small library!

ovos-core contains "the brains" of OpenVoiceOS, all the NLP components and skills are managed here

Skills Service

The skills service is responsible for loading skills and intent parsers

All user queries are handled by the skills service, you can think of it as OVOS's brain

All Mycroft Skills should work normally with ovos-core until version 0.1.0, after that modernization is required!

Under OpenVoiceOS skills are regular python packages, any installed skills will be loaded automatically by ovos-core

Since ovos-core 0.0.8 it is also possible to launch a skill standalone via ovos-workshop, this enables individual skill containers in ovos-docker

This can be also be helpful during skill development for quick testing before the skill is packaged

ovos-skill-launcher {skill_id} [path/to/my/skill_id]

Configuration

"skills": {

    // blacklisted skills to not load
    // NB: This is the skill_id, usually the basename() of the directory where the skill lives, so if
     // the skill you want to blacklist is in /usr/share/mycroft/skills/mycroft-alarm.mycroftai/
    // then you should write `["mycroft-alarm.mycroftai"]` below.
    "blacklisted_skills": [],

    // fallback skill configuration (see below)
    "fallbacks": {...},

    // converse stage configuration (see below)
    "converse": {...}

},

Utterance Transformers

NEW in ovos-core version 0.0.8, originally developed for Neon

when ovos-core receives a natural language query/ utterance from a user it is sent to a "preprocessing stage"

The utterance transformers framework consists of any number of plugins ordered by priority (developer defined), the utterance and message.context are sent sequentially to all transformer plugins, and can be mutated by any of those plugins

to enable a utterance transformer simply add it to mycroft.conf after installing it

// To enable a utterance transformer plugin just add it's name with any relevant config
// these plugins can mutate the utterance between STT and the Intent stage
// they may also modify message.context with metadata
// plugins only load if they are installed and enabled in this section
"utterance_transformers": {

 "ovos-utterance-normalizer": {},

 // cancel utterances mid command
 "ovos-utterance-plugin-cancel": {},

 // define utterance fixes via fuzzy match ~/.local/share/mycroft/corrections.json
 // define unconditional replacements at word level ~/.local/share/mycroft/word_corrections.json
 "ovos-utterance-corrections-plugin": {},

 // translation plugin
 "ovos-utterance-translation-plugin": {
   "bidirectional": true,
   "verify_lang": false,
   "ignore_invalid": true,
   "translate_secondary_langs": false
 }
},

Metadata Transformers

NEW in ovos-core version 0.0.8

Similar to utterance transformers, these plugins only transform the message.context

// To enable a metadata transformer plugin just add it's name with any relevant config
// these plugins can mutate the message.context between STT and the Intent stage
"metadata_transformers": {},

Pipelines

NEW in ovos-core version 0.0.8

after the utterance has been transformed it is sent to various OVOS components by priority order until one can handle the query

Pipelines include intent parsers, converse framework, common query framework and fallback skill framework

  // Intent Pipeline / plugins config
  "intents" : {
    // the pipeline is a ordered set of frameworks to send an utterance too
    // if one of the frameworks fails the next one is used, until an answer is found
    // NOTE: if padatious is not installed, it will be replaced with padacioso (much slower)
    // in the future these will become plugins, and new pipeline stages can be added by end users
    "pipeline": [
        "ocp_high",
        "stop_high",
        "converse",
        "padatious_high",
        "adapt_high",
        "fallback_high",
        "stop_medium",
        "adapt_medium",
        "ovos-persona-pipeline-plugin-high",
        "adapt_low",
        "common_qa",
        "fallback_medium",
        "ovos-persona-pipeline-plugin-low",
        "fallback_low"
    ]
  },

Pipeline Overview

NEW The concept of configurable pipelines has been introduced in ovos-core 0.0.8 and will be fully implemented as plugins after version 1.0.0

Most pipeline components will provide a high/medium/low variety with different confidence levels, but this is not required

pipeline	description	notes
stop_high	stop command exact matches	replaces OpenVoiceOS/skill-ovos-stop
converse	continuous conversation interception for skills
padacioso_high	high confidence intent matches from padacioso	slow !!! disabled by default
padatious_high	high confidence intent matches from padatious
adapt_high	high confidence intent matches from adapt
fallback_high	high priority fallback skill matches
ocp_high	high confidence media related queries
stop_medium	medium confidence stop matches	replaces OpenVoiceOS/skill-ovos-stop
padacioso_medium	medium confidence intent matches from padacioso	slow !!! disabled by default
padatious_medium	medium confidence intent matches from padatious
adapt_medium	medium confidence intent matches from adapt
ovos-persona-pipeline-plugin-high	active persona conversation (eg. a LLM)
ocp_medium	medium confidence media related queries
adapt_low	low confidence intent matches from adapt
common_query	send utterance to common_query skills and select best match
ocp_low	low confidence media related queries
fallback_medium	medium priority fallback skill matches
stop_low	low confidence stop matches	disabled by default
ovos-persona-pipeline-plugin-low	persona catch all fallback (eg. a LLM)
padacioso_low	low confidence intent matches from padacioso	slow !!! disabled by default
padatious_low	low confidence intent matches from padatious	almost always wrong disabled by default
fallback_low	low priority fallback skill matches