This documentation was last modified: Friday, April 5th, 2019 at 9:49 pm
Welcome to Mycroft.
As of late 2017, Mycroft is undergoing heavy development. It is at a level of maturity where developers and hardware hobbyists will be able to use it effectively. However, it is not yet ready for mainstream adoption.
This video provides a quick overview of our vision for Mycroft.
Mycroft is modular. Some components can be easily ‘swapped out’ for others:
- Wake Word detection
- Speech to Text (STT)
- Intent parser
A Wake Word is a phrase you use to tell Mycroft you’re about to issue a command. By default this is
Hey Mycroft, but you can configure your own Wake Word in your Mycroft Home account.
There are two technologies that Mycroft.AI currently uses for Wake Word detection:
- PocketSphinx: PocketSphinx is part of the broader CMUSphinx package, developed by Carnegie Mellon University. PocketSphinx is a lightweight speech recognition engine, specifically tuned for handheld and mobile devices.
Because PocketSphinx is trained on English speech, your Wake Word currently needs to be an English word, like
Hi there Mickey or
Hey Mike. Wake Words in other languages, like Spanish, French or German, won’t work as well.
- Precise: Unlike PocketSphinx, which is based on Speech to Text technology, Precise is a neural network that is trained on audio data. It doesn’t matter what words you want to use for your Wake Word. Instead, you train it on sounds. The downside is that Precise needs to be trained on your chosen Wake Word. Precise has been the default Wake Word Listener for each Mycroft Device since mid-March 2018. Prior to this date, PocketSphinx was the default.
Speech to Text (STT) software is used to take spoken words, and turn them into text phrases that can then be acted on.
Mycroft currently uses external Speech to Text software. However, we are aiming to build our own open Speech to Text solution, OpenSTT, by partnering with other organizations and their projects, such as Mozilla Voice.
- Google STT: Google STT is the default STT engine used with Mycroft. In home.mycroft.ai, you are able to change this default.
Mycroft also supports these other STT engines:
- IBM Watson Speech to Text (username and password required)
- wit.ai Speech to Text (wit.ai API key required)
An intent parser is software which identifies what the user’s intent is based on their speech. An intent parser usually takes the output of a Speech to Text (STT) engine as an input.
For example, Julie Speaks the following to Mycroft:
Hey Mycroft, tell me about the weather
Julie’s intent is to find out about the weather (probably in her current location).
An intent parser can then match the intent with a suitable Skill to handle the intent.
Adapt intent parser: Adapt is the default intent parser for all Mycroft platforms. Adapt was developed by Mycroft and is available under an open source license.
Padatious: Padatious is a neural network based intent parser. Padatious is currently under active development by Mycroft and is available under an open source license. It is likely that some Mycroft platforms will switch to using Padatious in the future instead of Adapt.
Text to Speech (TTS) software takes written text, such as as in text files on a computer, and uses a voice to speak the text. Text to Speech can have different voices, dependenting on the TTS engine used.
- Mimic: Mycroft’s default text to speech (TTS) engine, based on CMU’s Flite (Festival Lite)
In your home.mycroft.ai account, you can select several other TTS engines if you don’t wish to use Mimic.
- eSpeak: you need to choose which voice to use
- MaryTTS: you need to choose which voice to use
- Google TTS: you need to choose which voice to use
- FATTS: you need to choose which voice to use
The Mycroft middleware has two components:
Mycroft Core: this code, written in Python, is the core software that provides the ‘glue’ between other modules. Mycroft Core is available under an open source license.
Mycroft Home and Mycroft API: this is the platform where data on Users and Devices is held. This platform provides abstraction services, such as storing API keys that are used to access third-party services to provide Skill functionality. The code for this platform is not publicly available.
Mycroft Skills are like ‘add-ons’ or ‘plugins’ that provide additional functionality. Skills can be developed by Mycroft Developers, or by Community Developers, and vary in their functionality and maturity.
Mycroft Skills Management (msm) is a set of helper scripts that assist in the installation and management of Skills.
Mycroft is designed to run on many different platforms – called Devices, with different hardware, called Enclosures.
- Mark 1 – a software image of Mycroft designed to be installed on the Mycroft Mark 1 – a reference hardware device
- Picroft – a software image of Mycroft designed to be installed on Raspberry Pi 3
- Android – a software image of Mycroft designed to be installed on Android devices
Why would you choose to use Mycroft over alternatives such as Amazon™ Alexa™, Google Assistant™, Microsoft Cortana™ or Apple’s Siri™?
While alternative tools are readily available, and easy to use, they come with some pretty big downsides.
Mycroft is open source. This means that the code used by Mycroft can be inspected, copied, modified, and contributed back to the Mycroft community for everyone to enjoy.
Tools like Amazon™ Alexa™, Google Assistant™, Microsoft Cortana™ or Apple’s Siri™ are black boxes. That is, you can’t look inside them, and see what they do – or importantly – how they do it. And while they may allow you some flexibility to create new Skills, these are usually strictly controlled. Mycroft gives you freedom, flexibility and control over how your want your voice assistant to work. Mycroft is AI for everyone.
Mycroft code is licensed under the Apache 2 open source license.
Mycroft uses opt-in privacy. This means we will only record what you say to Mycroft with your explicit permission. Don’t want us to record your voice? No problem! If you’d like us to help Mycroft become more accurate, you can opt in to have your voice anonymously recorded.
Mycroft runs on a wide range of software platforms and hardware – meaning that you can run Mycroft on the Devices you want. Try Mycroft on a Raspberry Pi, on an Android device – or on your Linux desktop.
Mycroft is lightweight, and will run on low-powered hardware, such as the Raspberry Pi 3. So if you’re a Hacker, Maker or Hobbyist, it’s the perfect platform for low cost experimentation.
Mycroft has a vibrant, engaged and helpful community. You can interact with the community via: