[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: AUDITORY Digest - 11 Mar 2009 to 12 Mar 2009 (#2009-58)
On negative values of d'.
As Pierre Diveny says, d' is a distance, so it has no negative values. (Although I don't think Pierre is right in calling this a "bias" effect.) But of course people do sometimes operate below the diagonal in ROC space. In 3AFC it is tougher to interpret, although nasty things can happen with that method, since the signal is probably not detected equally often in each of the intervals. If it is not merely random variation, this interval effect may either be a bias or a real difference in detectability. A way to distinguish bias and d' effects is to run discrete control experiments, AAA vs AAB, AAA vs ABA, and AAA vs BAA. What a drag! In general it is good to remember that when the theoretical d' is 0.0, that half of the observed values will be below zero and half above.
BTW, we did the control experiment mentioned above. It is a tedious way to demonstrate the time order error.
Chuck Watson
Johnson, D.M., Watson, C.S. and Kelly, W.J. (1984). Performance differences among the intervals in forced choice tasks. Perception & Psychophysics, 35, 553-557.
-----Original Message-----
From: AUDITORY - Research in Auditory Perception [mailto:AUDITORY@xxxxxxxxxxxxxxx] On Behalf Of AUDITORY automatic digest system
Sent: Friday, March 13, 2009 12:14 AM
To: AUDITORY@xxxxxxxxxxxxxxx
Subject: AUDITORY Digest - 11 Mar 2009 to 12 Mar 2009 (#2009-58)
There are 6 messages totalling 480 lines in this issue.
Topics of the day:
1. HTK already optimized for TIMIT
2. Interpreting a negative d' (4)
3. [REMINDER] Sound and Music Computing Conference, Porto, Portugal, July
2009
----------------------------------------------------------------------
Date: Thu, 12 Mar 2009 07:08:55 +0000
From: Tony Robinson <tonyr@xxxxxxxxxxxxxxxxxx>
Subject: Re: HTK already optimized for TIMIT
> Date: Wed, 11 Mar 2009 21:25:30 +0100
> From: Javier Martinez Elicegui
>
> Dear all
>
> I am at the beginning of my thesis about applying human non-linear
> peripherical auditory models for speech recognition improvement in noisy
> environments.
>
> I have been preparing an ASR system with HTK toolkit and TIMIT corpus.
>
> Because I have some problems for optimizing HTK and this is not the focus of
> my thesis, I would be grateful if someone would send me a model
> configuration already optimized for TIMIT database (hmmdefs, macros,
> edfile.led,
You are in luck. Not quite as much luck as you asked for, but pretty lucky nevertheless.
I have written a bash script that trains and tests a TIMIT system for HTK. It's written for the purpose of research into acoustic features representations for automatic speech recognition, so you can just plug in your own features and it'll give you an error rate which tells you how good your are doing.
It can be downloaded from http://cantabresearch.dnsalias.com/HTKtimit.sh
Good luck,
Tony
Dr Tony Robinson
Founder, Cantab Research Ltd
------------------------------
Date: Thu, 12 Mar 2009 11:04:18 -0700
From: "Landsberger, David" <DLandsberger@xxxxxxx>
Subject: Interpreting a negative d'
I have conducted an experiment where I have obtained for one subject in one
condition a negative d' which I cannot explain. I was hoping that someone
here might be able to offer me some insight.
The experiment is a 3 interval forced choice task where a sound is presented
in each of the three intervals. In two of the intervals, the sounds are
identical. In the third interval, the sound is different. (Obviously the
order of stimuli is randomized.) The patient's task is simply to tell me
which of the three sounds is different.
In a 2IFC task, a negative d' might indicate that the subject has
misunderstood the task. For example, if two sounds were played and the
subject were asked to pick the sound that was higher pitched, a negative d'
would mean that the subject reliably picked the lower pitched sound as
having a higher pitched.
However, for a negative d' in a 3IFC task where two stimuli are the same,
the subject would have to reliably not pick the different one as being
different. What would that suggest about their perception? I don't think
its a misunderstanding of the task as the same subject was able to perform
fine in the same task using a different set of stimuli.
Any advice would be greatly appreciated.
Thanks,
David
------------------------------
Date: Thu, 12 Mar 2009 18:07:35 +0000
From: Luis Gustavo Martins <lgustavomartins@xxxxxxxxx>
Subject: [REMINDER] Sound and Music Computing Conference, Porto, Portugal, July 2009
--Apple-Mail-202-876017445
Content-Type: text/plain;
charset=ISO-8859-1;
format=flowed;
delsp=yes
Content-Transfer-Encoding: quoted-printable
[Sorry for cross-postings]
[Please circulate]
Call for Participation
Sound and Music Computing Conference, 23-25 July 2009
Casa da M=FAsica, Porto, Portugal
http://smc2009.smcnetwork.org/
SMC is a privileged forum in Europe for the promotion of international
exchanges around Sound and Music Computing topics. SMC 2009 will
feature papers, posters, music works, tutorials and special sessions.
=3D Important dates:
. Paper, Tutorials and Music submission deadline: Friday 17 April 2009
. [LATE DEADLINE] Deadline for participation to special sessions:
Monday 8 June 2009
Sound and Music Computing is an interdisciplinary field that
approaches the whole sound and music communication chain from a
multidisciplinary point of view.
By combining scientific, technological and artistic methodologies it
aims at understanding, modeling and generating sound and music through
computational approaches. Its core academic subjects relate to
musicology, physics (acoustics), engineering (including computer
science, signal processing and electronics),
psychology (including psychoacoustics, experimental psychology and
neurosciences) and music composition.
Topics will include:
. Automatic music generation/accompaniment systems
. Computer environments for sound/music processing/composition
. Networked music generation
. Physical modeling for sound generation
. Sound/music signal processing algorithms
. Digital Audio Effects
. Musical sound source separation and recognition
. Automatic music transcription
. Music information retrieval
. Musical pattern recognition/modeling
. Music and robotics
. Computational musicology
. Sonic interaction design
. 3D sound/music
. Data sonification
. Visualization of sound/music data
. Interfaces for music creation and fruition
. Interactive performance systems
. Musical performance modeling
. Gesture controlled audio systems
. Sound/music perception and cognition
. Sound/music and Neuroscience
. Music and emotions
. Multimodality in sound and music computing
. Web 2.0 and music
. Mobile music
. Technologies for the preservation, access and modelling of musical =20
heritage
All accepted papers and posters will be published in the conference
Proceedings, which will account for an ISBN. Furthermore, a special
issue of the Computer Music Journal will be devoted to articles based
on selected items from the conference. Not only will the best
conference papers be considered for publication
in extended form in the Journal, but also posters, tutorials, and
reports of special sessions.
=3D Note on Special Sessions:
=
http://smc2009.smcnetwork.org/call-for-participation/special-sessions.html=
SMC 2009 will feature special sessions on "hot" topics relevant to
Sound and Music Computing research, where a special focus will be put
on interaction between participants and the audience (members of the
scientific and music communities, and the general public at large).
There are, for now, 2 special sessions, on "Sonic Interaction Design"
and "Music Visualization".
http://smc2009.smcnetwork.org/programme/special-sessions/sid.html
=
http://smc2009.smcnetwork.org/programme/special-sessions/visualization.htm=
l
(Stay tuned for announces of more special sessions in the upcoming =20
weeks...)
=3D Note on Tutorials:
The day prior to the conference (22 July 2009) will consist of
tutorials, focusing on a single topic presented either at an
introductory level or in depth, lasting about 3 hours (including a
break).
You are welcome to propose tutorial along the guidelines detailed on
=
http://smc2009.smcnetwork.org/call-for-participation/call-for-tutorials.ht=
ml
=3D Stay tuned:
. RSS feed: http://smc2009.smcnetwork.org/news/news/RSS
. Facebook group: http://www.new.facebook.com/group.php?gid=3D42958077160
. Twitter: http://twitter.com/smc2009
=3D Want to help us promote SMC 2009?
. Follow and promote SMC 2009 on facebook and twitter
. Have a quick look here: http://smc2009.smcnetwork.org/promote-us
--Apple-Mail-202-876017445
Content-Type: text/html;
charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
<html><body style=3D"word-wrap: break-word; -webkit-nbsp-mode: space; =
-webkit-line-break: after-white-space; "><div style=3D"word-wrap: =
break-word; -webkit-nbsp-mode: space; -webkit-line-break: =
after-white-space; ">[Sorry for cross-postings]<br>[Please =
circulate]<br><br>Call for Participation<br><br>Sound and Music =
Computing Conference, 23-25 July 2009<br>Casa da M=FAsica, Porto, =
Portugal<br><a =
href=3D"http://smc2009.smcnetwork.org/";>http://smc2009.smcnetwork.org/</a>=
<br><br>SMC is a privileged forum in Europe for the promotion of =
international<br>exchanges around Sound and Music Computing topics. SMC =
2009 will<br>feature papers, posters, music works, tutorials and special =
sessions.<br><br><br>=3D Important dates:<br>. Paper, Tutorials and =
Music submission deadline: Friday 17 April 2009<br>. [LATE DEADLINE] =
Deadline for participation to special sessions:<br>Monday 8 June =
2009<br><br><br>Sound and Music Computing is an interdisciplinary field =
that<br>approaches the whole sound and music communication chain from =
a<br>multidisciplinary point of view.<br>By combining scientific, =
technological and artistic methodologies it<br>aims at understanding, =
modeling and generating sound and music through<br>computational =
approaches. Its core academic subjects relate to<br>musicology, physics =
(acoustics), engineering (including computer<br>science, signal =
processing and electronics),<br>psychology (including psychoacoustics, =
experimental psychology and<br>neurosciences) and music =
composition.<br><br>Topics will include:<br>. Automatic music =
generation/accompaniment systems<br>. Computer environments for =
sound/music processing/composition<br>. Networked music generation<br>. =
Physical modeling for sound generation<br>. Sound/music signal =
processing algorithms<br>. Digital Audio Effects<br>. Musical sound =
source separation and recognition<br>. Automatic music =
transcription<br>. Music information retrieval<br>. Musical pattern =
recognition/modeling<br>. Music and robotics<br>. Computational =
musicology<br>. Sonic interaction design<br>. 3D sound/music<br>. Data =
sonification<br>. Visualization of sound/music data<br>. Interfaces for =
music creation and fruition<br>. Interactive performance systems<br>. =
Musical performance modeling<br>. Gesture controlled audio systems<br>. =
Sound/music perception and cognition<br>. Sound/music and =
Neuroscience<br>. Music and emotions<br>. Multimodality in sound and =
music computing<br>. Web 2.0 and music<br>. Mobile music<br>. =
Technologies for the preservation, access and modelling of musical =
heritage<br><br>All accepted papers and posters will be published in the =
conference<br>Proceedings, which will account for an ISBN. Furthermore, =
a special<br>issue of the Computer Music Journal will be devoted to =
articles based<br>on selected items from the conference. Not only will =
the best<br>conference papers be considered for publication<br>in =
extended form in the Journal, but also posters, tutorials, =
and<br>reports of special sessions.<br><br><br>=3D Note on Special =
Sessions:<br><a =
href=3D"http://smc2009.smcnetwork.org/call-for-participation/special-sessi=
ons.html">http://smc2009.smcnetwork.org/call-for-participation/special-ses=
sions.html</a><br>SMC 2009 will feature special sessions on "hot" topics =
relevant to<br>Sound and Music Computing research, where a special focus =
will be put<br>on interaction between participants and the audience =
(members of the<br>scientific and music communities, and the general =
public at large).<br>There are, for now, 2 special sessions, on "Sonic =
Interaction Design"<br>and "Music Visualization".<br><a =
href=3D"http://smc2009.smcnetwork.org/programme/special-sessions/sid.html"=
>http://smc2009.smcnetwork.org/programme/special-sessions/sid.html</a><br>=
<a =
href=3D"http://smc2009.smcnetwork.org/programme/special-sessions/visualiza=
tion.html">http://smc2009.smcnetwork.org/programme/special-sessions/visual=
ization.html</a><br>(Stay tuned for announces of more special sessions =
in the upcoming weeks...)<br><br><br>=3D Note on Tutorials:<br>The day =
prior to the conference (22 July 2009) will consist of<br>tutorials, =
focusing on a single topic presented either at an<br>introductory level =
or in depth, lasting about 3 hours (including a<br>break).<br>You are =
welcome to propose tutorial along the guidelines detailed on<br><a =
href=3D"http://smc2009.smcnetwork.org/call-for-participation/call-for-tuto=
rials.html">http://smc2009.smcnetwork.org/call-for-participation/call-for-=
tutorials.html</a><br><br>=3D Stay tuned:<br>. RSS feed: <a =
href=3D"http://smc2009.smcnetwork.org/news/news/RSS";>http://smc2009.smcnet=
work.org/news/news/RSS</a><br>. Facebook group: <a =
href=3D"http://www.new.facebook.com/group.php?gid=3D42958077160";>http://ww=
w.new.facebook.com/group.php?gid=3D42958077160</a><br>. Twitter: <a =
href=3D"http://twitter.com/smc2009";>http://twitter.com/smc2009</a><br><br>=
=3D Want to help us promote SMC 2009?<br>. Follow and promote SMC 2009 =
on facebook and twitter<br>. Have a quick look here: <a =
href=3D"http://smc2009.smcnetwork.org/promote-us";>http://smc2009.smcnetwor=
k.org/promote-us</a><br><br><div apple-content-edited=3D"true"><span =
class=3D"Apple-style-span" style=3D"border-collapse: separate; color: =
rgb(0, 0, 0); font-family: Helvetica; font-size: 14px; font-style: =
normal; font-variant: normal; font-weight: normal; letter-spacing: =
normal; line-height: normal; orphans: 2; text-indent: 0px; =
text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; =
-webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: =
0px; -webkit-text-decorations-in-effect: none; -webkit-text-size-adjust: =
auto; -webkit-text-stroke-width: 0px; "><div style=3D"word-wrap: =
break-word; -webkit-nbsp-mode: space; -webkit-line-break: =
after-white-space; "></div></span></div></div></body></html>=
--Apple-Mail-202-876017445--
------------------------------
Date: Thu, 12 Mar 2009 15:14:40 -0400
From: Jon Boley <jdboley@xxxxxxxxxx>
Subject: Re: Interpreting a negative d'
If you are convinced that your subject understood the task correctly,
I would try to rule out these possibilities:
- Was the negative d' simply a matter of insufficient sampling?
(i.e., you randomly got an answer that looked significant) =A0It would
take more trials to determine if this is a likely explanation.
- Was there, in fact, some difference between the 2 "same" signals
that the listener picked up on, and you were not aware of? =A0Depending
on your test setup, this could be an artifact of the system, and it
may have only been audible for that stimulus (and/or for that
listener).
Jon
On Thu, Mar 12, 2009 at 2:04 PM, Landsberger, David
<DLandsberger@xxxxxxx> wrote:
>
> I have conducted an experiment where I have obtained for one subject in o=
ne
> condition a negative d' which I cannot explain. I was hoping that someone
> here might be able to offer me some insight.
>
> The experiment is a 3 interval forced choice task where a sound is presen=
ted
> in each of the three intervals. =A0In two of the intervals, the sounds ar=
e
> identical. =A0In the third interval, the sound is different. (Obviously t=
he
> order of stimuli is randomized.) =A0The patient's task is simply to tell =
me
> which of the three sounds is different.
>
> In a 2IFC task, a negative d' might indicate that the subject has
> misunderstood the task. =A0For example, if two sounds were played and the
> subject were asked to pick the sound that was higher pitched, a negative =
d'
> would mean that the subject reliably picked the lower pitched sound as
> having a higher pitched.
>
> However, for a negative d' in a 3IFC task where two stimuli are the same,
> the subject would have to reliably not pick the different one as being
> different. What would that suggest about their perception? =A0I don't thi=
nk
> its a misunderstanding of the task as the same subject was able to perfor=
m
> fine in the same task using a different set of stimuli.
>
> Any advice would be greatly appreciated.
>
> Thanks,
> David
------------------------------
Date: Thu, 12 Mar 2009 14:09:55 -0700
From: Pierre Divenyi <pdivenyi@xxxxxxxxx>
Subject: Re: Interpreting a negative d'
I remember that some 30-odd years ago I presented data at an
Acoustical Society meeting in which I showed upside-down responses in
a 2IFC task in form of negative d'. And I have never forgotten David
Green's (very kindly expressed) comment: d' always measures
discriminability which is a distance and which, by definition, can
never be negative. Therefore, every instance of a d' < 0 can only be
assigned to the observer's bias. So the question is not how to
interpret a negative d' but how to interpret the bias.
-Pierre
At 11:04 AM 3/12/2009, Landsberger, David wrote:
>I have conducted an experiment where I have obtained for one subject in one
>condition a negative d' which I cannot explain. I was hoping that someone
>here might be able to offer me some insight.
>
>The experiment is a 3 interval forced choice task where a sound is presented
>in each of the three intervals. In two of the intervals, the sounds are
>identical. In the third interval, the sound is different. (Obviously the
>order of stimuli is randomized.) The patient's task is simply to tell me
>which of the three sounds is different.
>
>In a 2IFC task, a negative d' might indicate that the subject has
>misunderstood the task. For example, if two sounds were played and the
>subject were asked to pick the sound that was higher pitched, a negative d'
>would mean that the subject reliably picked the lower pitched sound as
>having a higher pitched.
>
>However, for a negative d' in a 3IFC task where two stimuli are the same,
>the subject would have to reliably not pick the different one as being
>different. What would that suggest about their perception? I don't think
>its a misunderstanding of the task as the same subject was able to perform
>fine in the same task using a different set of stimuli.
>
>Any advice would be greatly appreciated.
>
>Thanks,
>David
------------------------------
Date: Thu, 12 Mar 2009 21:06:08 -0400
From: Robert Zatorre <robert.zatorre@xxxxxxxxx>
Subject: Re: Interpreting a negative d'
There is one other interpretation, which is not related to bias at all.
Rather, it is possible that the individual has indeed discriminated the
stimuli and is purposefully choosing the incorrect response. One can
only speculate about why someone would do this...it could be a sign of
malingering, psychosis, passive agression, or just plain orneriness!
Robert
-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Robert J. Zatorre, Ph.D.
Montreal Neurological Institute
3801 University St.
Montreal, QC Canada H3A 2B4
phone: 1-514-398-8903
fax: 1-514-398-1338
e-mail: robert.zatorre@xxxxxxxxx
web site: www.zlab.mcgill.ca
Pierre Divenyi wrote:
> I remember that some 30-odd years ago I presented data at an Acoustical
> Society meeting in which I showed upside-down responses in a 2IFC task
> in form of negative d'. And I have never forgotten David Green's (very
> kindly expressed) comment: d' always measures discriminability which is
> a distance and which, by definition, can never be negative. Therefore,
> every instance of a d' < 0 can only be assigned to the observer's bias.
> So the question is not how to interpret a negative d' but how to
> interpret the bias.
>
> -Pierre
>
> At 11:04 AM 3/12/2009, Landsberger, David wrote:
>> I have conducted an experiment where I have obtained for one subject
>> in one
>> condition a negative d' which I cannot explain. I was hoping that someone
>> here might be able to offer me some insight.
>>
>> The experiment is a 3 interval forced choice task where a sound is
>> presented
>> in each of the three intervals. In two of the intervals, the sounds are
>> identical. In the third interval, the sound is different. (Obviously the
>> order of stimuli is randomized.) The patient's task is simply to tell me
>> which of the three sounds is different.
>>
>> In a 2IFC task, a negative d' might indicate that the subject has
>> misunderstood the task. For example, if two sounds were played and the
>> subject were asked to pick the sound that was higher pitched, a
>> negative d'
>> would mean that the subject reliably picked the lower pitched sound as
>> having a higher pitched.
>>
>> However, for a negative d' in a 3IFC task where two stimuli are the same,
>> the subject would have to reliably not pick the different one as being
>> different. What would that suggest about their perception? I don't think
>> its a misunderstanding of the task as the same subject was able to
>> perform
>> fine in the same task using a different set of stimuli.
>>
>> Any advice would be greatly appreciated.
>>
>> Thanks,
>> David
------------------------------
End of AUDITORY Digest - 11 Mar 2009 to 12 Mar 2009 (#2009-58)
**************************************************************