timit

Documentation

Overview

Package timit Phones of the TIMIT database. For recognition testing the full set of 61 is typically reduced to 39 with confusable sounds folded into a group, e.g. "sh" and "zh" See Speaker-Independent Phone Recognition Using Hidden Markov Models, Kai-Fu Lee and Hsiao-Wuen Hon in IEEE Transactions on Acoustics, Speech and Signal Processing, Vol 37, 1989 for the original set and collapsing to 39 phones Many later studies use the 39 phone set

Index

• Variables
• func IdxFmSnd(s string, id string) (v int, ok bool)
• func LoadText(fn string) (string, error)
• func LoadTimes(fn string, names []string) ([]speech.Unit, error)
• func LoadTranscription(fn string) ([]string, error)
• func SndFmIdx(idx int, id string) (phone string, ok bool)

Variables

var PhoneCats10 = []string{"ah", "ao", "dh", "er", "ix", "iy", "l", "n", "r", "s"}

the PhoneCats10 set is a subset of phones that early results showed were more easily recognized and the set was used to "begin with success!"

var PhoneCats41 = []string{"iy", "ih", "eh", "ae", "ix", "ah", "uw", "uh", "ao", "ey", "ay",
	"oy", "aw", "ow", "l", "r", "y", "w", "er", "m", "n", "ng", "ch", "jh", "dh", "b", "d",
	"dx", "g", "p", "t", "k", "z", "zh", "v", "f", "th", "s", "hh", "pcl", "q"}

PhoneCats41 is a reduced set of phones. Some phones are generally recognized as confusable and swaps are not considered errors in the published phone recognition experiments. Some research does the mapping of many to one (e.g. n, nx and en all to "n") when scoring the recognizer output and the PhoneCats61 set is used. Using the PhoneCats41 set the mapping down is done during training and test.

var PhoneCats61 = []string{"iy", "ih", "eh", "ae", "ix", "ah", "ax", "ax-h", "uw", "ux", "uh", "ao", "aa", "ey",
	"ay", "oy", "aw", "ow", "l", "el", "r", "y", "w", "er", "axr", "m", "em", "n", "nx", "en", "ng",
	"eng", "ch", "jh", "dh", "b", "d", "dx", "g", "p", "t", "k", "z", "zh", "v", "f", "th", "s", "sh",
	"hh", "hv", "pcl", "tcl", "kcl", "bcl", "dcl", "gcl", "epi", "h#", "pau", "q"}

This is the full set of timit transcribed phones. See Lee & Hon, 1989 - Speaker-Independent Phone Recognition Using Hidden Markov Models

var Phones10 = map[string]int{
	"ah": 0,
	"ao": 1,
	"dh": 2,
	"er": 3,
	"ix": 4,
	"iy": 5,
	"l":  6,
	"n":  7,
	"r":  8,
	"s":  9,
}

var Phones41 = map[string]int{
	"iy":   0,
	"ih":   1,
	"eh":   2,
	"ae":   3,
	"ix":   4,
	"ah":   5,
	"ax":   5,
	"ax-h": 5,
	"uw":   6,
	"ux":   6,
	"uh":   7,
	"ao":   8,
	"aa":   8,
	"ey":   9,
	"ay":   10,
	"oy":   11,
	"aw":   12,
	"ow":   13,
	"l":    14,
	"el":   14,
	"r":    15,
	"y":    16,
	"w":    17,
	"er":   18,
	"axr":  18,
	"m":    19,
	"em":   19,
	"n":    20,
	"nx":   20,
	"en":   20,
	"ng":   21,
	"eng":  21,
	"ch":   22,
	"jh":   23,
	"dh":   24,
	"b":    25,
	"d":    26,
	"dx":   27,
	"g":    28,
	"p":    29,
	"t":    30,
	"k":    31,
	"z":    32,
	"zh":   33,
	"sh":   33,
	"v":    34,
	"f":    35,
	"th":   36,
	"s":    37,
	"hh":   38,
	"hv":   38,
	"pcl":  39,
	"tcl":  39,
	"kcl":  39,
	"bcl":  39,
	"dcl":  39,
	"gcl":  39,
	"h#":   39,
	"pau":  39,
	"epi":  39,
	"q":    40,
}

var Phones61 = map[string]int{
	"iy":   0,
	"ih":   1,
	"eh":   2,
	"ae":   3,
	"ix":   4,
	"ah":   5,
	"ax":   6,
	"ax-h": 7,
	"uw":   8,
	"ux":   9,
	"uh":   10,
	"ao":   11,
	"aa":   12,
	"ey":   13,
	"ay":   14,
	"oy":   15,
	"aw":   16,
	"ow":   17,
	"l":    18,
	"el":   19,
	"r":    20,
	"y":    21,
	"w":    22,
	"er":   23,
	"axr":  24,
	"m":    25,
	"em":   26,
	"n":    27,
	"nx":   28,
	"en":   29,
	"ng":   30,
	"eng":  31,
	"ch":   32,
	"jh":   33,
	"dh":   34,
	"b":    35,
	"d":    36,
	"dx":   37,
	"g":    38,
	"p":    39,
	"t":    40,
	"k":    41,
	"z":    42,
	"zh":   43,
	"v":    44,
	"f":    45,
	"th":   46,
	"s":    47,
	"sh":   48,
	"hh":   49,
	"hv":   50,
	"pcl":  51,
	"tcl":  52,
	"kcl":  53,
	"bcl":  54,
	"dcl":  55,
	"gcl":  56,
	"epi":  57,
	"h#":   58,
	"pau":  59,
	"q":    60,
}

Functions

func IdxFmSnd

func IdxFmSnd(s string, id string) (v int, ok bool)

IdxFmSnd returns the slice index of the snd if found. id is ignored if the corpus doesn't have subsets of sounds

func LoadText

func LoadText(fn string) (string, error)

LoadText retrieves the full text of the timit transcription

func LoadTimes

func LoadTimes(fn string, names []string) ([]speech.Unit, error)

LoadTimes loads both the timing and transcription data for timit files so the names slice is unused

func LoadTranscription

func LoadTranscription(fn string) ([]string, error)

LoadTranscription is a "no op" for timit, LoadTimes does the work of both

func SndFmIdx

func SndFmIdx(idx int, id string) (phone string, ok bool)

SndFmIdx returns the sound if found in the map of sounds of the corpus. id is ignored if the corpus doesn't have subsets of sounds