marf-bugs Mailing List for Modular Audio Recognition Framework

On Thu, 19 May 2016, Van Phu Tran wrote:

> Date: Thu, 19 May 2016 03:19:47 +0400
> 
> i am using your lib MARF for speaker recognition on eclipse java project it
> worked normal,

Good to know!

> but on Android have any bug like photo.
> please help me.
> sorry for my english not well.

I'll check your screenshot a bit later and will let you know what I can 
find. I know previously someone successfully ran in on Android before.

-- 
Serguei A. Mokhov, PhD                                   |  /~\     The ASCII
Computer Science and Software Engineering &              |  \ /  Ribbon Campaign
Concordia Institute for Information Systems Engineering  |   X     Against HTML
Concordia University, Montreal, Quebec, Canada           |  / \       Email!

Hello Hung,

On Thu, 4 Sep 2014, Hung Phan wrote:

> Date: Thu, 4 Sep 2014 04:49:39 +0700
> 
> Hello Professor Mokhov.
>
> My name is Hung D PHAN. Currently I am a researcher in Hanoi University of
> Science and Technology, Vietnam. One of my favourite research is speech
> processing.
>
> After time researching from the internet, luckily I found MARF is a good
> toolkit for voice feature extraction in java. In my task, I need to extract
> the MFCC information of input sounds. However, I can't find any
> documentation about MARF's MFCC extraction.
>
> *Sir, do you think extracting MFCC by MARF possibly? If yes, can you send
> my a further documentation which contains information about the
> configuration and approach for embedding code ?*

Adding MFCC extraction is certainly possible; as-is, however, it currently 
does not exist. But it should not be too difficult to add as a module. For 
our speaker identification we got away without MFCCs, using just FFT or 
LPC coefficients.

You'd need to add a class in marf.FeatureExtraction, say call it MFCC, see 
the class FFT as an example there. In the MFCC class's feature 
extraction method you'd need:

1. do FFT (we already have that, so simply call it)
2. remap the powers from FFT onto the mel scale by triangular overlapping
    windows (need to implement)
3. apply log() to the values obtained in 2. (need to implement)
4. take DCT of 3 (the mel log powers); DCT is like FFT but without the
    imaginary numbers so can re-use some of the FFT code we have again.

The output of 4 are your MFCCs that should be placed into the feature 
vector.

You'd have teach the main MARF class and FeatureExtractionFactory that the 
new module is available, and if you are using one of the apps, e.g., 
SpeakerIdentApp in the repo, it should be added with an option like --mfcc 
with the supporting code like the other options do.

I personally am completely out of time to do it myself at the moment, but 
you are welcome to contribute the module for the next version of MARF! :)

The best would be to check out the HEAD code from MARF's CVS.

> Thank you. If you want some information about my project, let visit my
> brief introduction at
> https://www.youtube.com/watch?v=-MZ-pXhvur0&list=UUzXgdewKO6F_KR-g6F9vPlw
>
> Thank you very much

Thanks, I'll have a look when I can.

Good luck!

-- 
Serguei A. Mokhov, PhD                                   |  /~\     The ASCII
Computer Science and Software Engineering &              |  \ /  Ribbon Campaign
Concordia Institute for Information Systems Engineering  |   X     Against HTML
Concordia University, Montreal, Quebec, Canada           |  / \       Email!

Bugs item #3432300, was opened at 2011-11-02 08:17
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Math
Group: MARF (any version)
>Status: Closed
>Resolution: Fixed
Priority: 5
Private: No
Submitted By: https://www.google.com/accounts ()
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Bad phase angle calculation in normalFFT method

Initial Comment:
Original code:
padPhaseAngle[i] = Math.atan(adOutputImag[i] / adOutputReal[i]);
Should be:
padPhaseAngle[i] = Math.atan2(adOutputImag[i], adOutputReal[i]);
Explanation:
viz wiki Complex number part Absolute value and argument

----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2011-11-17 10:08

Message:
Hi, thanks for the report. I could credit you by name or username if I had
one, but Google's OpenID only provided the URL. Yes, we've been only
calculating atan(y/x) case of the phase angle when x>0, and neglected the
other cases per http://en.wikipedia.org/wiki/Atan2. Fix committed. Will be
a part of whatever next release is, in the mean time please use the CVS
version.

Thanks,

-s

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Bugs item #3432300, was opened at 2011-11-02 08:17
Message generated for change (Settings changed) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Math
Group: MARF (any version)
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: https://www.google.com/accounts ()
>Assigned to: Serguei A. Mokhov (mokhov)
Summary: Bad phase angle calculation in normalFFT method

Initial Comment:
Original code:
padPhaseAngle[i] = Math.atan(adOutputImag[i] / adOutputReal[i]);
Should be:
padPhaseAngle[i] = Math.atan2(adOutputImag[i], adOutputReal[i]);
Explanation:
viz wiki Complex number part Absolute value and argument

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Bugs item #3432300, was opened at 2011-11-02 15:17
Message generated for change (Tracker Item Submitted) made by 
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Math
Group: MARF (any version)
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: https://www.google.com/accounts ()
Assigned to: Nobody/Anonymous (nobody)
Summary: Bad phase angle calculation in normalFFT method

Initial Comment:
Original code:
padPhaseAngle[i] = Math.atan(adOutputImag[i] / adOutputReal[i]);
Should be:
padPhaseAngle[i] = Math.atan2(adOutputImag[i], adOutputReal[i]);
Explanation:
viz wiki Complex number part Absolute value and argument

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=3432300&group_id=63118

Bugs item #1556320, was opened at 2006-09-11 08:37
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1556320&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Feature Extraction
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
>Assigned to: Serguei A. Mokhov (mokhov)
Summary: FeatureExtractionAggregator is broken: NullPointerException

Initial Comment:
Config: -norm -aggr -cheb
Mon Sep 11 07:58:46 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.792u 0.144s 0:01.94 99.4%     0+0k 0+0io 1pf+0w
Config: -norm -aggr -mah
Mon Sep 11 07:58:48 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.708u 0.108s 0:01.57 114.6%    0+0k 0+0io 1pf+0w
Config: -norm -aggr -randcl
Mon Sep 11 07:58:49 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.792u 0.144s 0:01.58 122.1%    0+0k 0+0io 1pf+0w


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-10-15 19:25

Message:
Logged In: YES 
user_id=560391

Fixed.

http://marf.cvs.sourceforge.net/marf/marf/src/marf/FeatureExtraction/FeatureExtractionAggregator.java?r1=1.7&r2=1.8&diff_format=u

Since the introduction of the new API of
extractFeatures(double[]), handling of its parameter was not
done properly when either querying a Preprocessing module or
acting directly on the parameter was reversed, and, also a
null in one case. Make it correct and for the parameter wrap
it into Raw, so the handling is always consistent with a
non-null Preprocessing. In the process of debugging this,
enhance error reporting.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1556320&group_id=63118

Bugs item #1568848, was opened at 2006-10-01 14:35
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1568848&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
>Assigned to: Serguei A. Mokhov (mokhov)
Summary: RandomClassification isn't producing 2 results throwing expt

Initial Comment:
This is during normal full-out run. Apparently,
there are also no training set file dumped with
RandomClassification...

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Config: -norm -fft -randcl
Sat Sep 30 14:28:25 EDT 2006
=============================================
No arguments have been specified.
1
java.lang.ArrayIndexOutOfBoundsException: 1
        at
marf.Storage.ResultSet.getSecondMininumID(ResultSet.java:127)
        at
marf.Storage.ResultSet.getSecondClosestID(ResultSet.java:169)
        at SpeakerIdentApp.ident(SpeakerIdentApp.java:593)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:385)


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-10-07 15:24

Message:
Logged In: YES 
user_id=560391

In response to the bug# 1568848, re-introduce explicit
dump() and restore() with proper redirection of the DUMP
modes back to the original as Classification alters them
for the training set. Additionally, prevent null oIDs to be
present when is read from the file. This fixes the bug and
makes RandomClassification correct again.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1568848&group_id=63118

Bugs item #1568848, was opened at 2006-10-01 14:35
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1568848&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Nobody/Anonymous (nobody)
Summary: RandomClassification isn't producing 2 results throwing expt

Initial Comment:
This is during normal full-out run. Apparently,
there are also no training set file dumped with
RandomClassification...

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Config: -norm -fft -randcl
Sat Sep 30 14:28:25 EDT 2006
=============================================
No arguments have been specified.
1
java.lang.ArrayIndexOutOfBoundsException: 1
        at
marf.Storage.ResultSet.getSecondMininumID(ResultSet.java:127)
        at
marf.Storage.ResultSet.getSecondClosestID(ResultSet.java:169)
        at SpeakerIdentApp.ident(SpeakerIdentApp.java:593)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:385)


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1568848&group_id=63118

Bugs item #1556320, was opened at 2006-09-11 08:37
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1556320&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Feature Extraction
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Nobody/Anonymous (nobody)
Summary: FeatureExtractionAggregator is broken: NullPointerException

Initial Comment:
Config: -norm -aggr -cheb
Mon Sep 11 07:58:46 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.792u 0.144s 0:01.94 99.4%     0+0k 0+0io 1pf+0w
Config: -norm -aggr -mah
Mon Sep 11 07:58:48 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.708u 0.108s 0:01.57 114.6%    0+0k 0+0io 1pf+0w
Config: -norm -aggr -randcl
Mon Sep 11 07:58:49 EDT 2006
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
There were errors in one or more aggregated modules:
[java.lang.NullPointerException,
java.lang.NullPointerException]
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeaturesImplementation(FeatureExtractionAggregator.java:262)
        at
marf.FeatureExtraction.FeatureExtractionAggregator.extractFeatures(FeatureExtractionAggregator.java:122)
        at
marf.MARF.startRecognitionPipeline(MARF.java:1404)
        at marf.MARF.train(MARF.java:1299)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:652)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:430)
1.792u 0.144s 0:01.58 122.1%    0+0k 0+0io 1pf+0w


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1556320&group_id=63118

Bugs item #1543529, was opened at 2006-08-20 12:21
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1543529&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Utilities
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Arrays.arrayToDelimitedString()s don't cope w/ empty arrays

Initial Comment:
The API family of Arrays.arrayToDelimitedString()
does not properly cope with the non-null arrays
of zero length.


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-09 13:28

Message:
Logged In: YES 
user_id=560391

The bug was corrected. Patch:

http://marf.cvs.sourceforge.net/marf/marf/src/marf/util/Arrays.java?r1=1.37&r2=1.38&diff_format=u

Will appear in the next CVS snapshot release.

-s


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1543529&group_id=63118

Bugs item #1541308, was opened at 2006-08-16 09:54
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1541308&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Storage Management
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: marf.Storage.Sample is not really Serializable

Initial Comment:
The Sample class cannot be serialized (to disk or
over the network) because its member
MARFAudioFileFormat has an oType member of
type AudioFileFormat.Type, which is not Serializable
and is not marked as transient causing
ObjectOutputStream to fail.


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-07 00:11

Message:
Logged In: YES 
user_id=560391

This bug has been corrected in the following patches:

http://marf.cvs.sourceforge.net/marf/marf/src/marf/Storage/MARFAudioFileFormat.java?r1=1.11&r2=1.12&diff_format=u
http://marf.cvs.sourceforge.net/marf/marf/src/marf/Storage/Sample.java?r1=1.44&r2=1.45&diff_format=u

Additionally, a JUnit test case for Sample was added
to test the Serializability (marf.junit.Storage.SampleTest).

-s


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1541308&group_id=63118

Bugs item #1551592, was opened at 2006-09-03 15:31
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1551592&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Applications - ZipfLawApp
Group: ZipfLawApp
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: ZipfLaw chokes on a single filename argument

Initial Comment:
The single filename argument does not work
as advertised when no other options are present
in ZipfLaw, e.g. by running 'make self':

...
ZipfLaw build ok
java -ea -verify -classpath marf.jar:. ZipfLaw ZipfLaw.java
Invalid options found: [ZipfLaw.java]
Usage:
    java ZipfLaw --help | -h | [ OPTIONS ] <corpus-file>
    java ZipfLaw --list [ OPTIONS ] <corpus-file>

Options (one or more of the following):
    --case  - make it case-sensitive
    --num   - parse numerical values
    --quote - consider quotes and count quoted strings
as one token
    --eos   - make typical ends of sentences (<?>, <!>,
<.>) significant
    --nolog - dump Zipf's law graph values as-is
instead of log/log
    --list  - lists already pre-collected statistics
for a given corpus

make: *** [self] Error 1

----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-03 17:02

Message:
Logged In: YES 
user_id=560391

The case when just a corpus filename is supplied was
overlooked in the option processing and was giving and
error of invalid options (when the total of ivalid and
submitted options count were "one"). The fix is committed
along with the ArrayIndexOutOfBounds one for this app.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1551592&group_id=63118

Bugs item #1459461, was opened at 2006-03-27 12:15
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1459461&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Applications - ZipfLawApp
Group: ZipfLawApp
>Status: Closed
>Resolution: Fixed
Priority: 7
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: irregular ArrayIndexOutOfBounds in aiFrequencies[f - 1]

Initial Comment:
There happens ArrayIndexOutOfBounds on index values
100,101,102 in line:

  aiFrequencies[iCurrFrequency - 1] = 1;

for a total of 190 times.


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-03 16:58

Message:
Logged In: YES 
user_id=560391

This bug was due to the fact that for the large corpora (and
depending on the tokenizer settings) there exist words with
the occurence frequency greater than 100, but we were
collecting up to the hundred only, so this was for now fixed
by placing a warning that beyond the page boundary, no C(f,w)
is computed. This ought to change for the general case, but
in some near future.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1459461&group_id=63118

Bugs item #1551592, was opened at 2006-09-03 15:31
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1551592&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Applications - ZipfLawApp
Group: ZipfLawApp
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: ZipfLaw chokes on a single filename argument

Initial Comment:
The single filename argument does not work
as advertised when no other options are present
in ZipfLaw, e.g. by running 'make self':

...
ZipfLaw build ok
java -ea -verify -classpath marf.jar:. ZipfLaw ZipfLaw.java
Invalid options found: [ZipfLaw.java]
Usage:
    java ZipfLaw --help | -h | [ OPTIONS ] <corpus-file>
    java ZipfLaw --list [ OPTIONS ] <corpus-file>

Options (one or more of the following):
    --case  - make it case-sensitive
    --num   - parse numerical values
    --quote - consider quotes and count quoted strings
as one token
    --eos   - make typical ends of sentences (<?>, <!>,
<.>) significant
    --nolog - dump Zipf's law graph values as-is
instead of log/log
    --list  - lists already pre-collected statistics
for a given corpus

make: *** [self] Error 1

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1551592&group_id=63118

Bugs item #1539695, was opened at 2006-08-13 18:03
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1539695&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: NullPointerException with the new train/classify(dbl[]) API

Initial Comment:
The training set filename generation in Classifcation
relies on the absolute presence of the non-null
reference to a feature extraction module, which may
not necessarily be true if one of the methods of
the new API (train(double[]) or classify([])) called
directoly, in this case getTrainingSetFilename()
throws a NullPointerException.


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-03 14:26

Message:
Logged In: YES 
user_id=560391

First the problem was spotted and fixed in the Distributed
MARF, and released in its PoC-demo, and now the fix made it
to the main tree. A patch has been committed that remedies
this problem by always keeping a local reference to the
feature array regardless where it is coming from.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1539695&group_id=63118

Bugs item #1550183, was opened at 2006-08-31 18:21
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1550183&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Preprocessing
Group: MARF (any version)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Endpoint class does not do normalization

Initial Comment:
While providing greater recongition performance for all
configurations that use
marf.Preprocessing.Endpoint.Endpoint, the class doesn't
use normalization, which is is not strictly fair.


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-09-02 12:03

Message:
Logged In: YES 
user_id=560391

In the just committed code along with the implemented
silence removal, normalization was added to Endpoint.
The recognition performance did not suffer for the top
result, which is a good thing :-).

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1550183&group_id=63118

Bugs item #1550183, was opened at 2006-08-31 18:21
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1550183&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Preprocessing
Group: MARF (any version)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Endpoint class does not do normalization

Initial Comment:
While providing greater recongition performance for all
configurations that use
marf.Preprocessing.Endpoint.Endpoint, the class doesn't
use normalization, which is is not strictly fair.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1550183&group_id=63118

Bugs item #1543529, was opened at 2006-08-20 12:21
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1543529&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Utilities
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Arrays.arrayToDelimitedString()s don't cope w/ empty arrays

Initial Comment:
The API family of Arrays.arrayToDelimitedString()
does not properly cope with the non-null arrays
of zero length.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1543529&group_id=63118

Bugs item #1541308, was opened at 2006-08-16 09:54
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1541308&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Storage Management
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: marf.Storage.Sample is not really Serializable

Initial Comment:
The Sample class cannot be serialized (to disk or
over the network) because its member
MARFAudioFileFormat has an oType member of
type AudioFileFormat.Type, which is not Serializable
and is not marked as transient causing
ObjectOutputStream to fail.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1541308&group_id=63118

Bugs item #1539695, was opened at 2006-08-13 18:03
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1539695&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: NullPointerException with the new train/classify(dbl[]) API

Initial Comment:
The training set filename generation in Classifcation
relies on the absolute presence of the non-null
reference to a feature extraction module, which may
not necessarily be true if one of the methods of
the new API (train(double[]) or classify([])) called
directoly, in this case getTrainingSetFilename()
throws a NullPointerException.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1539695&group_id=63118

Bugs item #1496654, was opened at 2006-05-28 21:48
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1496654&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Other
Group: MARF (CVS tip)
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: build.xml is broken for NetBeans and ant builds

Initial Comment:
As it's missing included build-impl.xml. Need
to replace with valid contents.

-s

----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-08-11 20:41

Message:
Logged In: YES 
user_id=560391

The NetBeans files were repaired and the version that
includes these can be downloaded from the link below.
Unfortunately, there are some "platform-specificity"
in them, and the ant under Linux doesn't work with
the current build.xml.

Download:

https://sourceforge.net/project/showfiles.php?group_id=63118&package_id=197804&release_id=438891


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1496654&group_id=63118

Bugs item #1459475, was opened at 2006-03-27 12:40
Message generated for change (Settings changed) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1459475&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Applications - Prob.ParsingApp
Group: ProbabilisticParsingApp
Status: Open
Resolution: None
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
>Summary: MARF keywords are recognized when parsing English

Initial Comment:
Some tokens are recognized as keywords of the MARF
language like 'not', 'or', 'and' per gram.lex.scan.log
instead of being bare words. May not be of a serious
problem though.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1459475&group_id=63118

Bugs item #1515258, was opened at 2006-06-30 13:47
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1515258&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: Classification.restore() calls dump() the the default case!

Initial Comment:
The default implementation of the
Classification.restore() calls super.dump()
intead of super.restore(). The error remained
unnoticed primarily because that the default
cases were not used that often.

-s

----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-07-28 13:17

Message:
Logged In: YES 
user_id=560391

This was corrected in the recent commit (well it was corrected
long ago but was not comitted yet). A similar problem was
fixed in the NeuralNetwor. The fix will appear in today's
CVS Tip release.

-s

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1515258&group_id=63118

Bugs item #1508630, was opened at 2006-06-19 09:40
Message generated for change (Comment added) made by mokhov
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1508630&group_id=63118

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: MARF - Classification
Group: MARF (CVS tip)
>Status: Closed
>Resolution: Fixed
Priority: 5
Submitted By: Serguei A. Mokhov (mokhov)
Assigned to: Serguei A. Mokhov (mokhov)
Summary: NeuralNetwork troubles

Initial Comment:
As of CVS tip NeuralNetwork dumps an empty
XML net (just the <net></net> tags) and the
dump files are missing a period in the filiname
prior the 'xml' extension:

marf.Classification.NeuralNetwork.NeuralNetwork.109.300xml


----------------------------------------------------------------------

>Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-07-28 13:15

Message:
Logged In: YES 
user_id=560391

A combined refactoring and fixes of Classification,
NeuralNetwork, and TrainingSet resolve the issues
mentioned in this ticket. NeuralNetwork and TrainingSet
are able to be dumped now in the DUMP_GZIP_BINARY
and DUMP_BINARY modes without problems and the whole
processes is a lot faster and takes less disk space.
The changes will appear in today's CVS Tip release.

----------------------------------------------------------------------

Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-06-30 12:50

Message:
Logged In: YES 
user_id=560391

Some further (little) debugging revealed that now Classification
forces the same format for the TrainingSet files as the
main classification module has it set. In this case,
NeuralNetwork uses XML by default, but the XML output is
not yet supported for TrainingSet, thus making it fail
leaving an empty net behind:

...
Not implemented: TrainingSet.restore() -- DUMP_BINARY, _XML,
_HTML, and _SQL are not supported yet.
        at
marf.Storage.TrainingSet.restore(TrainingSet.java:340)
        at
marf.Classification.Classification.loadTrainingSet(Classification.java:287)
        at
marf.Classification.Classification.train(Classification.java:128)
        at
marf.Classification.NeuralNetwork.NeuralNetwork.train(NeuralNetwork.java:198)
        at marf.MARF.train(MARF.java:1144)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:629)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:407)
Not implemented: TrainingSet.restore() -- DUMP_BINARY, _XML,
_HTML, and _SQL are not supported yet.
        at
marf.Classification.Classification.loadTrainingSet(Classification.java:305)
        at
marf.Classification.Classification.train(Classification.java:128)
        at
marf.Classification.NeuralNetwork.NeuralNetwork.train(NeuralNetwork.java:198)
        at marf.MARF.train(MARF.java:1144)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:629)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:407)
Phase: [restoring training set]
Phase: [restoring training set]
        at
marf.Classification.Classification.train(Classification.java:166)
        at
marf.Classification.NeuralNetwork.NeuralNetwork.train(NeuralNetwork.java:198)
        at marf.MARF.train(MARF.java:1144)
        at SpeakerIdentApp.train(SpeakerIdentApp.java:629)
        at SpeakerIdentApp.main(SpeakerIdentApp.java:407)
...

Previously, these two concepts were disjoint and
TrainingSet has always assumed DUMP_GZIP_BINARY, which
was indepedent of NeuralNetwork's DUMP_XML.

As a "by-product", TrainingSet seems to duplicate
the functionality for DUMP_GZIP_BINARY of the
StorageManager and does not implement just plain
DUMP_BINARY -- all this is to be refactored.



----------------------------------------------------------------------

Comment By: Serguei A. Mokhov (mokhov)
Date: 2006-06-25 12:40

Message:
Logged In: YES 
user_id=560391

The below is the fix for the extension problem:

http://marf.cvs.sourceforge.net/marf/marf/src/marf/Classification/NeuralNetwork/NeuralNetwork.java?r1=1.54&r2=1.55&diff_format=u

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=502900&aid=1508630&group_id=63118