biggestt / toney Goto Github PK

The sums used for variances and covarianses could be calculated in the following way by using a JS library for matrices. Since these calculations are done roughly 60 times per second and have a high level of complexity they might be a good subject for future optimisation.

The first time the soundfile plays is works but when trying to replay it either doesn't play, does play but no line or it plays with line drawing. Might have to do with the ugly setTimeout hack in toneModel initialize function

Using Chrome Devtools heap snapshot tool it can be seen that the application while running gathers up increasing amount of data that doesn't gets garbage collected. Addy Osmanis post helped finding this issue.
http://addyosmani.com/blog/taming-the-unicorn-easing-javascript-memory-profiling-in-devtools/

Snapshot of the heapsnapshots constant increase:

Currently there is only a threshold value in frequency spectrun variance that decides if a sample is noise or speech. Options to look into:

• Noise gate: http://en.wikipedia.org/wiki/Noise_gate
• Slider for current frequency variance threshold value
• Analyse data from microphone and find characteristics of the noise.

With docco the comments can be interpreted as markdown and nice documents like the one for the javascript tuner will be produced.

To get the third tone the regression analysis needs to be segmented. The breakpoint could be for example when the lines k-value changes sign or when a confidence test gets below a certain threshold value.

http://en.wikipedia.org/wiki/Lateral_inhibition

Try adjust the parameters of the HPS algorithm so that it outputs a better toneline.

Currently the hack to solve the audioended playback bug, #1 , involves reloading the soundfile and resetting the soundfileInputNode.

Since we only are interested in the relative change in tone over time the microphone and the user's tonelines dont have to be plotted on the same scale.

Set up a demo of the current development version of Toney on the gh-pages.