Sebastian Musslick

C

ode

Computational Modeling

Feedforward Task Model

In this 3-layer feedforward network stimulus information get's passed to the hidden layer and from there to the output layer. Both the hidden and output layer receive input from a task (control) layer that allows to modulate processing of the stimulus information (see Cohen, Dunbar & McCelland, 1990).

The model can be trained with backpropagation and optional weight decay.

Matlab Code

Reference:
Musslick S., Dey B., Özcimder K., Patwary M., Willke T. L., Cohen J. D. (2016). Controlled vs. Automatic Processing: A Graph-Theoretic Approach to the Analysis of Serial vs. Parallel Processing in Neural Network Archi- tectures. Proceedings of the 38th Annual Meeting of the Cognitive Science Society.

Experiment Design

Matlab Interface for IBM Watson Speech Recognition

Automatic speech recognition can be a useful tool for adaptive experiment designs or efficient post-hoc data analysis. This Matlab code provides an interface for the cloud-based Speech Recognition System Watson .

The code features the following functionality:

  • Online word recognition (see onlineClassificationDemo.m)
  • Post-hoc word recognition (see post-hocClassificationDemo.m)
  • Word data generation (see generateClassificationData.m)
  • Improved word classification based on similar keyword lists
  • Additional recognition outputs, such as of reaction time, word duration, classifier confidence and processing time


In order to use the interface, you need to acquire credentials for Watson services (the first thousand minutes of speech recognition data of every new month are free). Once the you set up your user credentials in speechRecogSettings.m, you can simply use the function recognizeWord(filepath) to classify any audio file. Please take a look at the README.txt for further detail.


Confusion matrix for a color word classification performed by Watson.