Talon script that allows you to record noise data while watching videos on
YouTube, Vimeo, etc. Just drop it somewhere in your Talon user
folder to use
it (or clone it as a git submodule).
The objective is to rapidly produce enough data to train noise recognition models as an alternative form of input. Here's an example of something similar using a different set of noises (the noises collected by Talon/this script avoid vocalization). Click to watch:
When you fullscreen a video, it will select a noise and start recording,
prompting you in the top-left corner. Repeatedly make the noise, pausing for
around a second each time (like you would on http://noise.talonvoice.com). The
noise will be recorded with all of your microphones at once to generate as much
data as possible, and saved in recordings/noises
in your .talon
folder
(talon
on Windows). When you exit fullscreen, the recording will stop.
In case you accidentally fullscreen a video, short recordings (less than 4 seconds) are ignored. Long recordings are split into 5-minute increments, so you can make the noise for as long as you want.
Once you have some data, post on the Slack.
Currently works on YouTube, Vimeo and Twitch. Feel free to add more sites like
Netflix - just add them to the context
.
The script will always record the noise for which you have recorded the least data. If you would like to avoid specific noises, comment them out directly.
Each microphone may be provided under many sources, so it may be recorded multiple times. Each source generates its own folder, so for now just ignore/delete duplicates.
If multiple sources have identical names, only one will be recorded.