I have two wave files (16k, 16bit, raw)

####

a.pcm, b.pcm

####

Their transcriptions are in "in.mlf'

#### in.mlf ####

#!MLF!#

"*/a.lab"

A

.

"*/b.lab"

B

.

 

############

 

Then I try

####

> HCopy -C config -I in.mlf -i out.mlf a.pcm + b.pcm out.htk

####

This means I want to make a big file containing both features of a.pcm and b.pcm while the transcriptions are maintained.

So, "out.mlf" is expected like this

#### out.mlf: expected ####

#!MLF!#

"*/out.lab"

00000000 56333124 A

56333124 143733124 B

.

############

 

 

However, HCopy yields 

 

#### out.mlf: actual output ####

#!MLF!#

"*/out.lab"

A

143733124 143733124 B

.

############

 

 

 

the config I used is general

#### config ####

SOURCEKIND  = WAVEFORM

SOURCEFORMAT = NOHEAD

SOURCERATE = 625

TARGETKIND = MFCC_0_D_A_Z

TARGETRATE = 100000.0

SAVECOMPRESSED = T

SAVEWITHCRC = T

WINDOWSIZE = 250000.0

USEHAMMING = T

PREEMCOEF = 0.97

NUMCHANS = 26

CEPLIFTER = 22

NUMCEPS = 13

ENORMALISE = T

ZMEANSOURCE = T

USEPOWER = T

 

############

 

Is it a bug or is it a right result?

 

Please let me know.

 

Thank you

 

Ted

--- (Edited on 2/16/2011 5:06 am [GMT-0600] by Visitor) ---

Hi Ted,

I did not know that you could create mlf files using HCopy...I see from the HCopy help page that it accepts mlf files, but I am pretty sure it cannot not generate time stamps - it might use them somehow, but I have not looked at HCopy in a while...

In the VoxForge tutorial we use a 2 step process to create the MLF file using the HTK perl script "prompts2mlf", and using HCopy to create the feature sets. 

I think what you are trying to do is forced alignment, i.e. find the word or phoneme times in your speech input. 

I don't think your can do that with HCopy - which just converts from one audio format to another or to feature sets.  You need a decoder like HVite or Julius or Sphinx to do 'forced alignment' - see here for more info: Automated Audio Segmentation Using Forced Alignment (Draft).

Ken

--- (Edited on 2/16/2011 1:47 pm [GMT-0500] by kmaclean) ---