This implements a natural language user interface for MPD (Music Player Daemon). It provides a restful webservice via flask which expects a instruction as string and a userid. To run the webservice type:
./flask/startServer.sh ./service/parse.py
To run a request type:
./flask/sendRequest.sh "Play David Bowie."
/flask -> scripts to start server or send requests
/service -> provides actual service content where instructions are parsed and MPD-commands are used
/locust -> stress tests
/spacy_analysis -> some scripts to explore spacy-features and testing how to parse new instructions
Download with
git clone --recursive
or fetch the submodule afterwards with:
git submodule update --recursive --init
Install python and (at least on Ubuntu 17.10) curl:
sudo apt install python3 python3-pip curl
Install dependencies:
pip3 install flask
pip3 install expiringdict
pip3 install spacy
pip3 install python-mpd2
Download model:
python3 -m spacy download en_core_web_lg
Due to conflicting folder structures you'll have to modify service/speech_processing/speech_control/text_to_speech.py: change
from resources.supported_languages import TTSLanguageCommand
to
from ..resources.supported_languages import TTSLanguageCommand
Play.
Stop.
Pause.
Play David Bowie.
Play Heroes from David Bowie.
Play not David Bowie.
Play David Bowie, Iron Maiden or Five Finger Death Punch.
Play not David Bowie, but Iron Maiden.
Don't play David Bowie.
Playing David Bowie would be nice.
My grandmother wants me to play David Bowie.
Don't stop.
Don't pause.
Resume.
Don't resume.
Play rock music or electro house.
Continue.
Stop playing
Play a random song.
Play random song.
Play something.
Play next.
Next.
Play rock music like Heroes from David Bowie.
Play previous song.
Previous.
Previous song.
Next song.
Clear current playlist.
Repeat playlist.
Repeat song.
Update Database.
Conversations are possible, since there is an internal state for each user. The state expires after a specified duration f.e. 10 seconds.
"Play Oompa Loompa music."
>> Hmm.. I didn't get that. Should i play rock?
"Yes"
>> Ok. Here we go.
"Play Oompa Loompa music."
>> Hmm.. I didn't get that. Should i play rock?
"No"
>> Oh, ok.
"Play something."
>> Tell me an artist, song or gerne. Or do you want me to play a random song?
"David Bowie"
>> Ok. Have fun!
"Play something."
>> Tell me an artist, song or gerne. Or do you want me to play a random song?
"random song"
>> Ok. Have fun!
This service provides support for more then one client. A user is simply specified by its given userid parameter.
For testing without a MPD-server, you can import the mpd_provider_module_debug instead of mpd_provider_module.
Instead of MPD-instructions, you can see an output on console.
You can do a stress test based on locust with following commands:
pip3 install locustio
locust -f locust/locustfile.py
Navigate to http://127.0.0.1:8089 specify number of users, hatchrate and start.
Feel free to specify some test cases in the locustfile.py! In my tests, the service performs pretty well (on debug mode).
"Playing David Bowie would be terrible." seems to need a sentiment analysis like in http://nlp.stanford.edu:8080/sentiment/rntnDemo.html. -> But "Playing David Bowie, since this is a very bad day." would return negative result. In addition spaCy doesn't provide a pre-trained model. Sentiment analysis would help in a few cases but leads to misinterpretation of much clearer instructions. Therefore sentiment analysis is not used.