Constant Field Values


Contents
arch.*

arch.Polyphone
public static final java.lang.String SB "|"
public static final java.lang.String WB "#"

bin.*

bin.Converter
public static final short LABEL_SIZE 12
public static final java.lang.String SYNOPSIS "Translate between various file formats.\n\nusage: java bin.Converter in_format out_format < data_in > data_out\n\nformats:\n ufv,dim\n Unlabeled feature data, 4 byte (float) per sample dimension\n lfv,dim,label1,label2,...,labeln\n Labeled feature data; 12 byte label, then 4 byte (float) per sample.\n Label ID will be attached according to the sequence of labels in the\n argument: label1 -> 0, label2 -> 1, etc. as the Sample class requires\n numeric labels.\n frame\n Unlabeled feature data, 8 byte (double) per sample dimension\n sample\n Labeled feature data using the statistics.Sample class\n ascii\n Unlabeled ASCII data: TAB separated double values, one sample per line.\n ascii_label\n Labelled ASCII data: TAB separated values, first field is label.\n"

bin.GaussEM
public static final java.lang.String SYNOPSIS "Estimate Gaussian mixture densities using an initial estimate and a\n(large) data set.\n\nusage: java bin.GaussEM <options>\n -i initial-model\n Initial estimate of the mixture density. See bin.Initializer for\n possible starts.\n -n iterations\n Number of EM iterations to compute.\n -o output-model\n File to write the final estimate to.\n -l listfile\n Use a list file to specify the files to read from.\n -p num\n Parallelize the EM algorithm on num cores (threads). Use 0 for \n maximum available number of cores. NB: -p 1 is different from -s as\n it doesn\'t cache the entire data set.\n -s\n Do a standard single-core EM with a complete caching of the data.\n This might be faster than -p for small problems with less files.\n --save-partial-estimates\n Write out the current estimate after each iteration (to output-model.*)\n\ndefault: -n 10 -p 0\n"

bin.Initializer
public static final java.lang.String SYNOPSIS "Mixture initializer, sikoried 06/2009\nusage: java bin.Initializer strategy num-clusters {diagonal: true|false} outfile [-l list] [feature-file1]\n\nAvailable strategies:\n knn\n Find the clusters by iteratively distribute the data into the\n num-cluster clusters, refining the centroid in each step.\n\n g [strategy]\n Hierarchical, statistically driven Gaussian clustering, similar\n to the LBG algorithm.\n Available strategies:\n none : split cluster if not normally distributed (no re-ranking)\n cov : split the cluster with highest covariance\n sum_ev : split the cluster with the highest sum of eigen values of\n the covariance\n diff_ev : split the cluster with the highest difference in eigen\n values\n ad_score : split the cluster with the highest Anderson-Darling\n statistics\n ev : compare densities by the largest EV"

bin.Map
public static final java.lang.String SYNOPSIS "MAP adaption for mixture densities, bocklet & sikoried 07/2009\n\nAdapt an initial mixture density using the given feature data. If\nnum-iterations is specified, the MAP step is repeated.\n\nusage: java bin.Map -i <initial> -o <adapted> [-a adaptation-mode] [-r <relevance>] [-n num-iterations] [-l list] [-f file]\n\nadaptation-mode:\n \'p\' : update priors\n \'m\' : update priors\n \'c\' : update covariances\n\nDefault parameters: -a pmc -n 1 -r 16\n"

bin.RandomClusterGenerator
public static final java.lang.String SYNOPSIS "usage: java RandomClusterGeneratr density-string1 number1 [density-string2 number2...] > feature-file\n\ndensity-string: mue1,mue2,cov1,cov2"

bin.SuperVector
public static final java.lang.String SYNOPSIS "SuperVector generator, sikoried 07/2009\n\nGenerate super vectors from mixture densities and concatenate them to\na frame file.\n\nusage: java bin.SuperVector <param-string> [model1 ...] > frame-file\n param-string:\n \'p\' : include priors\n \'m\' : include means\n \'c\' : include (diagonal) covariances\n"

sampled.*

sampled.AudioCapture
public static final java.lang.String synopsis "usage: AudioCapture -r <sample-rate> [options]\nRecord audio data from an audio device and print it to stdout; supported\nsample rates: 16000, 8000\n\nOther options:\n -L\n List available mixers for audio capture and exit; the mixer name\n or ID can be used to specify the capture device (useful for\n multiple microphones or to enforce a certain device).\n -m <mixder-name>\n Use given mixer for audio input instead of default device\n -a\n Change output mode to ASCII (one sample per line) instead of SSG\n -o <out-file>\n Save output to given file (default: stdout)\n -h\n Display this help text\n"

statistics.*

statistics.Density
public static final double MIN_PROB 1.0E-50
public static final double REGULARIZER 1.0E-10

statistics.DensityFull
public static final transient double GAMMA 0.0010

statistics.MixtureDensity
public static final java.lang.String SYNOPSIS "usage: statistics.MixtureDensity mue,cov [mue,cov ...] > mixture\n\nCreate a mixture with equal priors and the given densities. mue and \ncov are comma separated lists of double values. DIAGONAL COVARIANCE ONLY!\n"

util.*

util.Gnuplot
public static final java.lang.String SYNOPSIS "usage: statistics.MixtureDensity data-file <details:none|id|details> mixture1 [mixture2 ...]"