Precise
Precise is a Wake Word Listener based on a neural network trained on sound samples.
Like its name suggests, a Wake Word Listener's job is to continually listen to sounds and speech around the Device, and activate when the sounds or speech match a Wake Word. Unlike other hotword detection products, Mycroft Precise is fully open source.
Precise is based on a neural network that is trained on sound patterns rather than word patterns. This reduces the dependence it has on particular languages or accents.
PocketSphinx is an alternative to Precise. Unlike Precise, PocketSphinx recognizes Wake Words based on the CMU Flite dictionary of sounds.
Precise is the default Wake Word Listener for the "Hey Mycroft" wake word, PocketSphinx provides a fallback to this if Precise is unavailable.
To change the Wake Word Listener to PocketSphinx if it has been set to Precise, Speak:
Hey Mycroft, set the Listener to default
or
Hey Mycroft, set the listener to PocketSphinx
Mycroft will respond
"I've set the Listener to PocketSphinx"To return to Precise, speak:
Hey Mycroft, set the Listener to Precise
Mycroft will respond
"I've set the Listener to Precise"To find out which Wake Word Listener is active for the Mycroft Device you are using, simply Speak:
Hey Mycroft, what is the Listener?
or
Hey Mycroft, tell me what Listener you are using
If you are using Precise, Mycroft will respond:
"The current Listener is Precise"If Precise is not already installed, speak to your Device:
Hey Mycroft, set the Listener to Precise
Mycroft will respond
"Downloading the new listener, this will take a bit and you won't be able to use me until it completes. Give me a minute before attempting to activate me again."NOTE: As Precise is installing, it will download a 40Mb file to your Device. During this initial download period, your Device will not be able to respond to a Wake Word.
Unlike other accurate hotword detection products, Mycroft Precise is fully open source.
| Open Source | Accurate | Languages | Model Trainer | Commercial Use | GitHub |
Precise | | | Python | Permitted | ||
Snowboy | | | Node, Java, Python, GO, Perl, iOS, Android | |||
Porcupine | | | C, Python, Android | |||
PocketSphinx | | | Almost all | Permitted |
Training your own custom Wake Word model for Precise requires at least functional experience using the Linux commandline and an understanding of basic machine learning concepts. It requires time and training data.
An instructional overview is available on the Precise repository on Github. Community member El-tocino has also provided a short write up of their tips for getting the best result.
If you have questions, you can reach out to others in the Community via the Troubleshooting channel in Chat.
Last modified 2yr ago