hugvey/README.md

Hugvey / Pillow Talk

• Panpoticon
  • Fancy nickname for the web interface that allows altering the story and running the individual Hugveys
• Voice
  • Lyrebird voice syntehsis API wrapper. Set the oAuth token using a token generated here
• Client
  • Individual Hugveys that stream their mic output and play audiofiles trough the Panopticon. Communication with the server is done through zmq
  • Connect with them trough hugvey1.local etc (1-25).
• Central Command/server
  • One server to rule them all. Start individual threads/subprocesses for the individual Hugveys. The Panopticon is started when starting the server.

Server

Run the server: python hugvey_server.py --config server_config.yml

To make sure: don't run the server in a screen subterminal, as the amount of output can clog the server

Screen commands:

Set scrollbuffer in ~/.screenrc: as described here

ctrl+a d: detach, ctrl+a esc: scroll the buffer, ctrl+a i: current buffer settings

Panopticon

The server also integrates the panopticon, the monitoring & administration interface to Hugvey.

Client

To run it: python hugvey_client.py -c client_config.yml

Development

The Panopticon uses gulp to compile SASS into CSS, and to set up browser-sync for css & js. For now, no js user facing dependencies are managed trough node/npm.

After starting the server:

cd www
gulp

To run a command on all hugveys:

fab -H rubenvandeven.com,saclab@projects.rubenvandeven.com -- uname

Installation

create and load Python virtualenv

virtualenv -p python3 --system-site-packages venv
source venv/bin/activate

Install requirements

apt install libasound-dev python3-pyaudio git-lfs
pip install -r requirements.server.txt
pip install -r requirements.txt

Install soft requirements (These are hard requirements for the clients!)

apt install sox rsync

Don't forget to init git submodules

git submodule init
git sumodule update

copy autossh-hugvey.service to /etc/systemd/system/:

cp autossh-hugvey.service /etc/systemd/system/autossh-hugvey.service
sudo systemctl enable autossh-hugvey.service 

clients

apt install supervisor

Add the following to the first section of /etc/supervisor/supervisord.conf

chown=pi:pi

Deploy / usefull commands

rsync them all

for i in {1..26}; do echo $i; rsync -av ~/hugvey/ pi@hugvey$i.local:/home/pi/hugvey/ --exclude=www --exclude=venv --exclude=local --exclude=*.pyc --exclude=.git --exclude=recordings --exclude=/voice* --exclude=/pd --exclude=Lanbox; done

shut all of them down

for i in {1..26}; do ssh pi@hugvey$i.local "sudo shutdown -h now"; done

restart supervisor on all of them

for i in {1..26}; do ssh pi@hugvey$i.local "supervisorctl restart hugvey_client"; done

install bash on all of them

for i in {1..26}; do echo $i;ssh pi@hugvey$i.local "cd hugvey && sudo bash install_server.sh"; done

make them say their number

for i in {1..26}; do echo $i;ssh pi@hugvey$i.local "espeak -f /etc/hostname"; done

Select wifi network:

for i in {1,2,3,6,7,8,12,13,17,18,20,22}; do echo $i;ssh pi@hugvey$i.local "/sbin/wpa_cli -i wlan0 select_network 0"; done

Monitoring the server processes

To make sure it will not die with 'Too many files open'

examine open file limit with ulimit -a and set it with ulimit -n NUMBER

lsof -p $(ps aux|grep "[h]ugvey_server.py" |awk '{print $2}')| awk '{print $9}'|sort -rn|uniq -c|sort -rn|head -20

or

lsof | grep $(ps aux|grep "[h]ugvey_server.py" |awk '{print $2}')| awk '{print $11}'|sort -rn|uniq -c|sort -rn|head -20

Branches

multichannel

Experiment with using alsa+sox to stream multiple channels to google.

Create Story

Messages

Things that can/will be said by Hugvey

Text

The text that will be said. Or just a description if custom audio is uploaded. Variables can be entered by predending them with a $dollar_sign.

Start

Mark message as being the start of a strand/tree of messages. Used for eg. diversions and formatting of the editor

Beginning

There can only be one beginning. This is the message that the Hugvey will start with when starting the story.

Chapter Start

A chapter can be marked. This is used by the timeout diversions, as it only returns to the next chapter marker after diversion. Furthermore, it is used by diversions to prevent them from triggering if specific sections (chapter markers) have been played.

Audio

Upload a custom audio file to override the auto generated file

Afterrun time

deprecated Was the time the microphone kept listening after triggering this message. It was used to have the Hugvey reconsider its direction. Not used anymore

Volume factor

Parameter send to the play command to increase/decrease the playback volume

Tempo factor

Parameter send to the play command to increase/decrease the playback speed

Pitch factor

Parameter send to the play command to increase/decrease the playback pitch (minus values for lower pitch)

Color

Color the message aids in finding it in the editor window. Also is used by replycontains diversions to prevent from running in specific moments

Directions

Directions connect messages from one to the other. Can be created in the editor by selecting a message, and shift+click on its follow-up message. Another way is to select a message and press 'create message' from the right bar. This new message will automatically be connected and inherit the same color.

Condition

Messages only head to a specific following message when one of the direction's conditions is matched (OR-condition). First come, first served.

The Conditions Description field allows for giving some info on the condition's reasons.

Types of conditions:

• timeout: timing finishing the playback of the message's audio
  • seconds: the duration
  • Only if no reply: timeout will be disabled after the person has spoken anything
  • Reply needed: If checked, the timeout is counted when it is met. This counter is used by the consecutive-timeouts diversions.
• replyContains: Match the contents of the speech using a regex. Or just any speech. Used to capture variables.
  • regex: The regex to match on. Variables can be matched using the python syntax to give the variable a name (?P<variable_name>\w+)
  • three consecutive timings can be given:
    • delay reply duration: the duration of the reply since hugvey stopped speaking. If it is more than this value (but less than the larger) it will use the given timing
    • delay wait time: The time to wait after a person speaks. It doesn't wait for Google's is_finished parameter, but rather checks from Google's last response. This way, also short utterances sutch as 'hey' or 'ok' are also properly timed, as these often don't get an is_finished by Google.
  • instant match: don't use any timings. the moment the regex matches on the speech in progress, the condition is met.
• variable: returns True if variable is set
  • TODO
• diversion: returns True if diversion has ben taken.
  • TODO

Diversions

• ReplyContains: Match the contents of speech using regex.
  • works only if the current message has one, and only one, direction.

times occured/only on n-th instance: determines the order of diversions of the same type (for Timeout and no_response). Starting at 1, as a diversion with value of 0 can occur always

Network

4G Modem

Visit 192.168.5.1 The password is at the bottom of the device.

Restoring a hugvey

• Write image /mnt/stash/hugvey.img to the microSD card.
  • You can use gnome-disks 'restore image' for that
• Open partition called rootfs, and sudo nano etc/hostname
  • Change hugvey20 into hugveyX (the number you need)
• Start it, rsync, install_server.sh as mentioned above.

Pure Data patch

The pure data patch controls the loop for both the python script and the max patch.

If there's a click in the audio (which randomly occurs), change audio settings. For some reason PD cannot remember/store these. So after each start of PD set Audio Settings:

Sample rate

44100 Block size

1024 Delay, msec

50