To start using DIVA
type: DIVA; at the Matlab command prompt.
This will initialize all model components in start a GUI for interfacing
with all model components.
The model is preset with a number of learned sounds. These sounds are displayed
in a menu on the GUI. Clicking one of these will load the relevant model state
variables. Produce the selected sound, click the "Produce Sound" button. Once
the model is finished producing this sound, you can play the sound (i.e. synthesize)
by pressing the "Play Sound" button.
To learn a new target
First browse for the target (MAT file) by clicking the "Browse
Button." This will select a target file. Next press the
"Acquire Target" button to input the sound to the DIVA model.
Finally, when the model is finished inputting the target, select
the target from the "Stored Auditory Targets" list and press
"Train".
To learn/relearn a previously acquired target
Select the desired sound from the "Stored Auditory Targets" list.
Then press "train" to continue learning from the last saved
point. Pressing "Reset Weights" reinitializes model state variables to
begin training an aquired target from scratch.
Changing model parameters
You can change model parameters by selecting the "Parameters" menu and then selecting
the individual parameter, or by selecting the "Edit All"
Changing application preferences
To change the number of training trials performed per "Train" action,
select the appropriate item from the "Preferences" menu.
Simulations involving "noisy" computation
Can be initiated by designing the additive noise distribution to corrupt the
following three computations:
- Acoustic error corrective movement
- Somatosensory error corrective movement
- Feedforward Command
Noise distribution design is accessable through the Experimental Control->Model Noise
menu item.
Batch learning of all targets in a specific directory
To learn all targts in Sessiontargets/Session_default/ folder (in batch mode),
type: LoadDIVA
at the Matlab command prompt.
This will initiate the DIVA model and commence learning of targets
in the Sessiontargets/Sessiondata/ folder. WARNING: This action
will take a VERY long time.