ERROR [+6550] LoadHTKLabels: Junk at end of HTK transcription
FATAL ERROR – Terminating program HResults

After inspecting the reference MLF file, I discovered a blank line in one of the transcriptions. After deleting this line, the error disappeared, and HResults completed successfully.

First, the Switchboard pronouncing dictionary can be downloaded here (along with the transcriptions).  The filename of the dictionary in this release is sw-ms98-dict.text.  Then, the CMU pronouncing dictionary can be downloaded here. For my purposes, I am disregarding the lexical stress information provided by the CMU dictionary, so I have removed all of the ’0′, ’1′, and ’2′ labels following the vowels.

If you compare the phones in the two dictionary files, you will see that the following three phones are present in the Switchboard dictionary, but absent in the CMU dictionary:  ax, el, and en. To convert the Switchboard dictionary transcriptions to match the CMU phone set, ax should be changed to AH, el should be changed to AH L, and en should be changed to AH N.

Here is a simple sed script that I wrote to apply these changes to the Switchboard dictionary file sw-ms98-dict.text (in addition to removing comments and empty lines):

$ sed '1 d' sw-ms98-dict.text \
| sed '/^#/ d' \
| sed '/^$/ d' \
| sed 's/ ax/ ah/g' \
| sed 's/ en/ ah n/g' \
| sed 's/ al/ ah l/g' \
| sed 's/ .*/\U&\E/'

Note that this sed script requires GNU sed in order to use the last command to change the replacement text to upper case.

I solved this by passing the following two arguments to the configure script:

$ configure --without-x --disable-hslab

After the new Makefiles were generated, make all worked and built all of the HTK tools except HSLab.

$ HCopy -C config -S train.scp

I received the following warning message for each file:

WARNING [-6371] ValidCodeParms: Using linear spectrum with PLP in HCopy

The feature vector I was attempting to use was defined in the configuration file to be:

TARGETKIND = PLP_0_D_A_Z

The warning message disappeared when I added the following line to the configuration file:

USEPOWER=T

The format of each line of the parameter file is:

dataType name = value;

The data type for each parameter should be available in the man page for the command. To run the command using these parameters, you can either simply name the parameter file params and put it in the directory where you run the command, or use the -P command line option to specify the name of the file. For example, the following command will use the parameter file f0params.txt to produce pitch estimates for the audio file filename.wav:

$ get_f0 -P f0params.txt filename.wav filename.f0

$ HLStats -b train.bigram -o train.wlist train/*.lab
Segmentation fault

where train.bigram is the filename where the language model should be saved, train.wlist contains a list of all of the unique words in the corpus, and train/ is a directory that stores a .lab orthographic transcription file for each speech file.

By turning the HTK verbosity to the highest level (-T 8) I was easily able to identify the file that caused the segmentation fault:

$ HLStats -T 8 -b train.bigram -o train.wlist train/*.lab
. . .
Processing file train/XXX.lab
Segmentation fault

I discovered that the file XXX.lab was empty: the transcriber did not transcribe anything because the speech in that file was inaudible. This was apparently the cause of the Segmentation Fault in HLStats. So, one solution would be to make sure all label files contain at least one transcribed word.

After glancing at the source code, though, in HLStats.c, it seemed like HLStats should be able to handle empty label files. Indeed, when I concatenate all of the label files into a Master Label File, HLStats works fine and prints a warning message for the empty label files:

$ HLStats -T 8 -b train.bigram -o train.wlist train.mlf
. . .
Processing file train/XXX.lab
WARNING [-1330] HLStats: Empty file XXX.lab in HLStats
. . .

So, it looks like HLStats isn’t able to process empty label files when they are listed individually as command line arguments.

$ HBuild wordList wordLoop

The word loop grammar simply contains a network in which each word occurs with equal probability.

To recognize a speech file using this word loop grammar, run the following command (assuming that macros and hmmdefs contain the acoustic models, dict contains the pronouncing dictionary, and monophones contains a list of the HMM names):

$ HVite -H macros -H hmmdefs -w wordLoop dict monophones testFile

This type of recognition is slow if there are more than a few words in wordList, and the accuracy is generally low. It is useful, however, for experimental purposes.

After compiling and installing SoX 14.3.0 on Ubuntu 9.04, I tried to convert an Ogg Vorbis file to WAV format, but received the following error:

$ sox filename.ogg filename.wav
sox FAIL formats: no handler for detected file type ‘vorbis’

Another sign that something was wrong was the following output from running ./configure before compiling SoX:

OPTIONAL FILE FORMATS
oggvorbis…………………….no

To fix this problem and enable the use of Ogg Vorbis files with SoX, I did the following:

$ sudo apt-get install vorbis-tools

In addition I downloaded and installed libogg and libvorbis from Xiph.Org (in that order, since libvorbis depends on libogg). After that, I re-configured SoX, and saw

OPTIONAL FILE FORMATS
oggvorbis…………………….yes

Then, after re-compiling and re-installing, SoX was able to process the Ogg Vorbis file.

After downloading the source code (version 14.3.0) and succesfully compiling it on Ubuntu 9.04, I received the following error when trying to start up SoX:

$ sox
sox: error while loading shared libraries: libsox.so.1: cannot open shared object file: No such file or directory

After a little poking around, I see that the file libsox.so.1 was installed to /usr/local/lib when SoX was installed. Also, /usr/local/lib is included in the file /etc/ld.so.conf (the system file which contains a list of shared libraries). However, to link these newly installed libraries and make them accessible, I needed to run the following command:

$ sudo ldconfig

After running that, SoX worked just fine.

>>> import re
>>> pat=re.compile(r’([A-Za-z]+)-([A-Za-z]+)’)
>>> s = ‘This is a te- test of back-references in Python.’
>>> re.sub(pat, r’\1 \2′, s)
‘This is a te- test of back references in Python.’