presentation

Presentation demo.
It recognizies the first phoneme of a spoken name.
Demonstrates recognition of names (when in range). Non-recognition of name when it is not a match, or if it is out of range.

Currently working on:
final report (sorry, writing always takes me a while).
finishing up log-sampling. The problem is more complicated than I thought. I realized that I don't actually know how to "graph" in matlab. Writing my own "spectrogram" function and sorting buckets require a different distribution and using controls that I'm slowly learning due to a lack of documentation. Will work on it after the final report.

following this example:

http://www.mathworks.com/matlabcentral/fileexchange/1553-spectrogram-short-time-ft-log-magnitude

I've been playing around with different sampling schemes.

Here's a derpy graph XD

Poster done

This poster took me longer than expected >x<

beta video.

Realized I forgot to blog last week. I'll make up for it once I get some sleep. C:


take 1

take 2

Reading and Researching

Caught up on a lot of the readings.

 I learned what dynamic time warping is.
I also learned contradictory things about sampling for sound.


Started coding the part for logrithmatic sampling, but decided a better spend of time may actually be to work on getting a good demo for next week. (My beta review is April 8th). So I've put aside the sampling code for now, and is working on getting the preceptor to understand their name.

After finishing that, i'll work on intergrating microsoft speech sdk into this whole thing.

If I'll have more time, then I'll go back and work on logrithmatic sampling. 

HCA Tree accomplished

The HCA tree is a weird form of huffman encoding.
The parent's weight is the average of the children weights.

Will look at how to sample over the weekend.

Before I sleep toda

Before I sleep today I will produce the HCA tree. I don't want to spend any more time being confused about which two nodes should be next to each other.

I'll be coding in Java, the algorithm will be similar to huffman encoding.

sudo code:
given confusion matrix, create nodes that represent each node's similarity. Use greedy approach (approximation) to construct the first pair, then proceed from there.

I don't particularly know if this'll work but I'll start from there.