Player is loading...
ESR09 : Clinical relevance of intelligibility mesures
Pommée Timothy
Wednesday, Sept. 4, 2019 · 4:49 p.m. · 12m 13s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
hello run uh first of all sorry for my broken voice input in german i was
00:00:05
and i just want to remind you that i'm a speech and language pathologist another
00:00:09
computer scientists so you will not hear about any complex models or stuff like that okay
00:00:15
the uh will um talk about what ah down uh during my first year in
00:00:20
to lose with the remote yeah you can point to and a dish and whatnot
00:00:26
um first of all i i hear about a
00:00:29
systematic review is um my colleague matter but i get
00:00:33
i'm so used to press my guidelines uh which
00:00:37
require a two raters use where we collaborated on this
00:00:41
uh for that we went to also to be trained by
00:00:44
purpose are uninspired on the methodology off the the prisoner guidelines
00:00:50
and uh the question of this review was ah what acoustic measures are
00:00:54
actually used uh in literature to um accounts for speech intelligibility help adults
00:01:02
and uh so i'm not gonna go into detail but um maybe kind twenty two studies
00:01:09
um and we know it it's a separate beings so first of all i
00:01:15
in all of the studies very um there is a high variability off
00:01:19
acoustic measures that i used uh to try to count per speech intelligibility
00:01:24
and also a high variability off a subjective uh that's a intelligibility estimates
00:01:30
i'm only five of all the studies uh directly investigated the
00:01:35
correlation between the acoustic measure and the uh proceed speech intelligibility
00:01:40
and um most of these correlations were either i'm not significant or um
00:01:46
the uh results were conflicting with their yeah we can do studies
00:01:51
um so three main issues where race it's uh in this uh
00:01:58
review the first one is that uh intelligibility is uh the find very um
00:02:05
yeah i in in many if a different ways in a uh in in the
00:02:09
literature so there is no consents a consensus
00:02:13
on what actually uh this uh term means
00:02:17
uh and so this will have an impact on um how do you feel uh
00:02:23
how scientists were uh try to uh assess it's what acoustic
00:02:27
measures they will use a hollow with what methodology they use
00:02:32
ah there is also a lack of credit given to the age
00:02:35
and gender factors so a lot of um the papers actually um
00:02:40
to all the uh either only ages or the genders together
00:02:45
and um based on the aim of the study uh the results are um
00:02:52
are reported very um well the methodology
00:02:56
uh and the results are rhythm reported uh
00:03:00
was uh sometimes incomplete in a in a complete manner sorry
00:03:06
um so there is still an extensive research needed to identify which acoustic
00:03:11
or let's uh i can be used to account for a precede speech intelligibility
00:03:16
and uh so this paper has been submitted to the uh john of us jenna
00:03:21
of speech language and hearing the search but was rejected so we have to um
00:03:25
review it and then try to resubmitted to a more appropriate i don't know
00:03:31
uh we also did this second to systematic review with semi true um
00:03:37
in partnership with directly project and the a research question was which acoustic
00:03:42
measures i used to say speech intelligibility this time in the ontological contacts
00:03:47
and it's uh that one what's actually except for um
00:03:52
it will soon be a published in the head and neck
00:03:56
an actual um then my rabbit next was a um apart from the
00:04:02
literature to see what uh the clinicians actually use in their current practised
00:04:07
so i used in the online survey uh it on um for spring speaking uh uh clinicians
00:04:14
i was born recipes and for fun iterations by only had the answers for missile piece
00:04:20
and the questions uh i had was what uh do they
00:04:23
actually use today in the clean the praxis to um assessed speech
00:04:28
and what are the needs and what are the lax off um existing tools
00:04:35
and so um well basically these are the the idea is that uh uh
00:04:42
that uh i came out of this uh service so they need a yeah um
00:04:49
and you don't think they need new tools and uh one of the most important uh
00:04:55
aspects what is new to this subjectivity so um as a way to the underlined it
00:05:02
uh the there is a problem is actually a subjective measures uh which are not reliable
00:05:09
and um so we found like basically the conclusions of
00:05:13
of this sort of you are similar to be uh the
00:05:16
ones in uh that were found five religions coming or an
00:05:20
f. four b. um clinical practised in the u. s. a.
00:05:26
um and we also found that uh most of the clinicians are unfamiliar with
00:05:31
us uh objective to actually so they're not used at all by most of them
00:05:37
and um they're not a familiar with us
00:05:40
uh acoustic parameters or recording parameters so um
00:05:44
there is a need for standardisation and simplification after recording crocheted your
00:05:50
um there is also a need for clear instructions on how to
00:05:53
record the patients and uh also to have clear instructions for the task
00:05:58
because um they use very different tasks they use different
00:06:02
speech material and they don't really know which one is best
00:06:05
um they need a they ask for easy access to assessment
00:06:09
tools and also for training in the objective assessment of speech
00:06:14
um for the first ah appointed for d. and standardisation
00:06:18
after recording procedure we're currently i'm bringing together working group um
00:06:25
of people who uh for professionals to uh try to create a reference text
00:06:30
so that's a clinicians can use one text that is uh that has been constructed a
00:06:36
uh by a panel of speech expert that to take into
00:06:39
account the most stuff uh most of permit us a possible
00:06:46
um so i had the um literature review then uh uh the uh the online survey for the um
00:06:52
point of view of the conditions and to supplement this online survey i also
00:06:57
and try to have a is any guide interviews with speech experts um
00:07:04
which visitor basically tackles the same question so how do they assess speech intelligibility
00:07:09
and what are the lax of currently existing tools and i have already interviewed for
00:07:14
experts at the yeah the out of congress in august and uh a lot of
00:07:22
interviews to come
00:07:25
and uh and this is the third point that i have been in uh during
00:07:30
my first year so it it was first attend in the um acoustic analysis speech
00:07:35
so um this was inspired by the uh speech banana that is used in our geometry
00:07:42
we just basically um representation of the consonants on
00:07:46
me i ordered ram so you have this uh
00:07:50
banana and it represents the speech sounds in when you draw the um the auditory to uh
00:07:58
a curve you can see 'em which uh will you can estimate which speech
00:08:05
sounds are will be received or not received by the patient based on this graph
00:08:10
and so uh well the first question was how was this
00:08:13
banana uh actually uh created how big the plot is uh
00:08:19
uh continents on the graph and um so we try to uh also
00:08:24
do the same but in the product in um in the production no
00:08:29
yeah the production we um you we recorded a fifteen subjects knotty subject
00:08:34
um and we tried to uh identified by a spectral peaks um
00:08:41
yeah some examples off the words to plots these uh the consonants actually
00:08:47
on it to access graphs we have the um the intensity and frequencies
00:08:53
so as an example i have here the uh
00:08:56
that is where the recordings off a d. m. sounds
00:09:00
um by women so you can see that there are a recurrent peaks and we try to identify
00:09:07
these recurrent peeks into plot them on a disk laugh doesn't really look like a banana but uh
00:09:14
so um the goal of this this actually to have 'em required
00:09:18
global representation of a speech articulation by uh by patient and um
00:09:27
this was the pilot study so we know that there are we only
00:09:30
had fifteen subject so there is a need for a a way larger sample
00:09:35
and also we know that there are lots of limits to this a methodology
00:09:38
because we can not just applied this a peek a spectral peak uh speaking
00:09:43
for each phonemes for example for the fricatives it is not rational to only
00:09:48
use peaks because you will um use a frequency bands for example so um
00:09:57
uh_huh well basically there for next year the the project to try
00:10:01
to um continue on working on this uh acoustic to to um
00:10:07
supplemented with additional acoustic cues and a lot
00:10:12
of questions that are still uh open okay
00:10:16
um so the next steps forward that next year's
00:10:19
ah on for so we'll try to resubmit systematic review
00:10:24
uh we go uh first have to re modify it a bit and then um
00:10:29
right to submit a continually expert interviews
00:10:33
um and then on the medium term well
00:10:37
as i just said uh tried to continue developing this uh this speech banana to um
00:10:45
and then uh we'll all significant and um for that and not actually i'm
00:10:49
quite sure we talked about sheffield uh id is not there anymore in the um
00:10:57
so basically i'm just i'm still trying to find the and aim for this segment and because i don't want
00:11:03
just to go away without knowing why so um based
00:11:07
on my presentations if you have any ideas i'm open
00:11:12
and uh on the long term it would also i tried to find out how we can
00:11:16
take into account the functional impact off to speech impairments on the patients ah quality of life
00:11:23
and um for that i also collaborate a lot was
00:11:26
a mature works on this um functional impacts a perspective
00:11:33
so as a summary of my eyes and the contributions there has been a five abstract
00:11:40
and that posters and presentations at conferences a
00:11:44
a one a systematic review that has been accepted
00:11:47
and uh i also submitted in abstract for a next year um at the motor speech conference
00:11:56
and santa barbara
00:12:00
and uh this is a summary of my first year for a while
00:12:07
do you have any questions
00:12:09
hi
Conference Program
Storage and management of sensitive speech data
Henk van den Heuvel
Sept. 4, 2019 · 11:51 a.m.
ESR01 : Speech based automatic dementia detection in a home environment
Yilin Pan
Sept. 4, 2019 · 2:14 p.m.
ESR03 : Interpretable speech pathology detection
Julian Fritsch
Sept. 4, 2019 · 2:30 p.m.
ESR04 : Acoustic speech based monitoring for telehealth and predictive analytics
V. Srikanth Nallanthighal
Sept. 4, 2019 · 2:54 p.m.
160 views
ESR07 : Modeling progression of patients with neurodegenerative disorders
Camilio Vásquez
Sept. 4, 2019 · 4:05 p.m.
ESR08 : Developing measurment procedure of pathological speech intelligibility
Wei Xue
Sept. 4, 2019 · 4:31 p.m.
ESR09 : Clinical relevance of intelligibility mesures
Pommée Timothy
Sept. 4, 2019 · 4:49 p.m.
Big Data with Health Data
Sébastien Déjean
Sept. 5, 2019 · 9:20 a.m.
Small Data, Number of features vs. number of observations
Olivier Crouzet
Sept. 5, 2019 · 9:47 a.m.
ESR11 : First year review
Bence Halpern
Sept. 5, 2019 · 11:20 a.m.
ESR10 : Acoustic features to support the perceptual evaluation of accent production in dysarthric speech
Viviana Mendoza Ramos
Sept. 5, 2019 · 11:39 a.m.
110 views
ESR12 : Continuous speech recognition for people with dysarthria
Zhengjun Yue
Sept. 5, 2019 · 11:50 a.m.
153 views
Management of health data in an industrial company
V. Srikanth Nallanthighal
Sept. 5, 2019 · 1:57 p.m.
146 views
