A helper class for handling audiodb databases
load features from wildcard search into vectorized covariance stacked matrix, one file per row
read time-series features from a file list, convert to covariance vectors and stack as a matrix
turn time-series into Gaussian parameter vector consisting of mean and vectorized covariance
data - multidimensional feature array
mn - use means, True/False [True]
cv - use covariances, True/False [True]
if mn==False and cv==False, use raw data
num_blocks - divide data into blocks, must be an integral divisor of rows [1]
deltas - use derivatives, True/False [True]
num_frames - number of frames to keep, 0=all [0]
norm_type - vector norming, 0=none, 1=L1, 2=L2, 3=infinity [0]
post_norm - apply norming to output vectors (in addition to data vectors) [0]
frame_off - offset start of data by frame_off [0]
Returns:
a vector corresponding to requested features
read time-series features from a file and convert to vectorized means and covariance matrix
Retrieve ADB database, or create if doesn't exist
Place features X and powers P into the adb database with unique identifier given by string "Key"
Simple insert features into an audioDB database named by dbBame.
Features are either chroma [default], mfcc, or cqft.
Feature parameters are default.
Walk the list of features, powers, keys, and, optionally, times, and insert into database
read a binary little-endien row-major adb array from a file.
Uses open, seek, fread
read directory of binary features with wilcards, e.g. dir('Features/*/*.mfcc')
returns a list of observation matrices, one for each file in the wildcard search
read feature data from a list of filenames into a list of arrays
resample the columns of data by a factor of prop e.g. 0.75, 1.25,...)
Static search method
returns sorted list of results
Sort search results by stripping out repeated results and placing in increasing order of distance.
Sparseness measure of row vector data.
Returns a value between 0.0 (smooth) and 1.0 (impulse)
Static tempo-invariant search
Returns search results for query resampled over a range of tempos.
Remove repeated results from result list
write a binary little-endien row-major adb array to a file.