Haiku Deck is a presentation tool that allows users to create Haiku style slideshows. The tool encourages users to get started making their own Haiku Deck presentations which can be shared on 際際滷Share. In just a few sentences, it pitches the idea of using Haiku Deck to easily create brief, visually focused presentations.
1) El documento describe el ciclo de Carnot y el teorema de Clausius, dos conceptos fundamentales de la termodin叩mica. 2) El ciclo de Carnot es un ciclo reversible ideal que consta de dos procesos isot辿rmicos y dos procesos adiab叩ticos. 3) El teorema de Clausius establece matem叩ticamente que la suma algebraica de los cocientes entre la cantidad de calor y la temperatura es igual a cero para cualquier ciclo reversible.
Ecuador es una rep炭blica ubicada en el noroeste de Am辿rica del Sur, limitando con Colombia y Per炭, con una extensi坦n de 283,561 km2. Su capital y ciudad m叩s grande es Quito, y su moneda oficial es el d坦lar estadounidense. El pa鱈s se divide en cuatro regiones principales: Costa, Sierra, Amazon鱈a y Gal叩pagos.
Dit is de powerpoint van onze bestuursrecruitering en info-drink van 2015-2016. Indien je vragen zou hebben, stuur ons dan gerust een mail via recruiting@academicsfordevelopment.be.
Ferrater mora. diccionario de filosofia. tomo i. (espanhol)JJusticiera
油
El autor resume los cambios realizados en la quinta edici坦n de su Diccionario de Filosof鱈a. Escribi坦 546 nuevos art鱈culos, reescribi坦 totalmente 213 art鱈culos, y ampli坦 o modific坦 262 art鱈culos. Adem叩s, revis坦 el texto, corrigi坦 errores, ampli坦 datos y actualiz坦 la bibliograf鱈a. Como resultado, esta edici坦n contiene una gran cantidad de material nuevo en todas las disciplinas filos坦ficas y en la historia de la filosof鱈a. Aunque trabaj坦 arduamente por ampliar y mejorar la obra, el
Curtis Miller is a Mechanical Engineer with over 24 years of experience in manufacturing and high-tech environments. He currently works as a Process Improvement Engineer at Curtis Products, Inc. where he promotes and implements process improvements, interprets technical documents for new product development, and develops internal processes. Previously, he worked as a Machine & Tool Design Engineer at Nibco, Inc. where he read blueprints and technical documents, completed mechanical designs, and recommended design modifications. He has expertise in Pro/E Wildfire and CREO 2.0 and holds an Associate of Science degree from Purdue University.
Este documento describe los conceptos fundamentales de la energ鱈a interna y el trabajo para gases ideales y no ideales. Explica que la energ鱈a interna de un gas depende de la temperatura y el volumen, y que aumentos en la energ鱈a cin辿tica se manifiestan como aumentos en la temperatura, mientras que aumentos en la energ鱈a potencial se manifiestan como aumentos en el volumen. Tambi辿n describe los diferentes tipos de procesos termodin叩micos como adiab叩ticos, isob叩ricos e isot辿rmicos.
ACRL 2015 Presentation: Models of Library Engagement with MOOCsLaura O'Brien
油
Panel discussion description:
Discover how libraries at national and international higher education institutions are engaging with the design, production, and implementation of MOOCs at their campuses. What strategies have libraries employed to find the time, money, and staff necessary to develop models of support for production of (and teaching in) MOOCs, in addition to other distance learning initiatives? What you learn may influence whether and how deeply your library engages with MOOCs at your own institution.
1) La iluminaci坦n es la medida del flujo luminoso incidente sobre una superficie. Se define como el flujo luminoso por unidad de 叩rea y se mide en lux. 2) La intensidad luminosa de una fuente es el flujo luminoso emitido por unidad de 叩ngulo s坦lido y se mide en candelas. 3) La iluminaci坦n de una superficie depende de la intensidad de la fuente de luz y est叩 inversamente relacionada al cuadrado de la distancia entre la fuente y la superficie.
This poem explores themes of confusion, inner conflict, and the struggle to find clarity or meaning through references to sinking, hidden tricks, and paradoxes. The speaker seems lost in their thoughts and unable to break free from an endless cycle of questions without answers that leave them feeling perplexed and confined in their own mind.
El presente proyecto tuvo como objetivo evaluar un Manual de Funciones Administrativas dirigido al Consejo Comunal Sector 4 Urbanizaci坦n La Carucie単a. Se realiz坦 mediante una investigaci坦n de campo descriptiva con una muestra de 51 voceros y voceras del Consejo Comunal, a quienes se les aplicaron un cuestionario y una encuesta para indagar sobre la aplicaci坦n de las funciones establecidas en el Manual. Los resultados mostraron que si bien los voceros conocen el Manual, su implementaci坦n es escasa, pues se observ坦 poca informaci坦n para ofrec
O documento discute como fraudes podem ocorrer tanto no setor p炭blico quanto no privado, contrariando a afirma巽達o de que "Fraudes s坦 ocorrem no setor p炭blico!". Ele fornece estat鱈sticas mostrando que 5% do PIB mundial 辿 perdido para fraudes e estima que no Brasil isto representaria R$207 bilh探es perdidos. Alerta as empresas que 辿 preciso estar preparado contra fraudes pois a oportunidade e a falta de preparo podem levar a sua ocorr棚ncia.
Este documento presenta conceptos fundamentales de luminotecnia, incluyendo las magnitudes y unidades utilizadas como flujo luminoso, intensidad, iluminaci坦n y brillo. Explica la ley de Weber-Fechner que rige la relaci坦n no lineal entre est鱈mulo y sensaci坦n para la vista y otros sentidos. Tambi辿n describe las caracter鱈sticas y tipos de fuentes de luz, y c坦mo var鱈an la intensidad y iluminaci坦n con la distancia y 叩ngulo de incidencia de los rayos de luz.
During the planning and research stage for her project, Ifrah Khalif used various media technologies. She conducted online research using YouTube to watch previous student work and film trailers in the horror genre. Google was also used for research of film posters and magazines. Her mobile phone helped with organization and communication. A blogger was created to document the process and share research sources like questionnaires and schedules. Adobe software like Photoshop, After Effects, and Premiere Pro were used for editing tasks like the film poster, production ident, and final video.
The document contains two stories. The first story is about a 100 year old tree on a university campus that has witnessed history, including soldiers assembling under it before leaving for World Wars 1 and 2. Experts say the tree is remarkably healthy for its age and should live another 50-75 years. The second story is about an experienced hiker, Will Henderson, who broke his leg after slipping in a stream off the Appalachian Trail. Despite being experienced, he hiked alone and got lost. He crawled for 4 days with a makeshift splint before being rescued. He plans to finish hiking the trail but will not hike alone off the main path again.
An overview of 'Step by Step', the final report for the Shared Decision Making in BC.
The Shared Decision Making (SDM) in British Columbia Project was a multi-year independent research initiative housed at Simon Fraser Universitys Centre for Dialogue. This collaborative project examined the emergence of non-treaty SDM Agreements, negotiated between British Columbia and individual or groups of First Nations, in the form of Strategic Engagement Agreements or Reconciliation Protocols. Each SDM Agreement provides a framework for collaboration between two governments who despite their differences, seek to build working relationships with one another, develop trust, and find ways to reach mutually agreeable decisions about how land and resources should be managed.
The overarching purpose of the SDM in BC project was to understand more clearly where SDM agreements have come from, what they mean, and how they are working so far.
Curtis Miller is a Mechanical Engineer with over 24 years of experience in manufacturing and high-tech environments. He currently works as a Process Improvement Engineer at Curtis Products, Inc. where he promotes and implements process improvements, interprets technical documents for new product development, and develops internal processes. Previously, he worked as a Machine & Tool Design Engineer at Nibco, Inc. where he read blueprints and technical documents, completed mechanical designs, and recommended design modifications. He has expertise in Pro/E Wildfire and CREO 2.0 and holds an Associate of Science degree from Purdue University.
Este documento describe los conceptos fundamentales de la energ鱈a interna y el trabajo para gases ideales y no ideales. Explica que la energ鱈a interna de un gas depende de la temperatura y el volumen, y que aumentos en la energ鱈a cin辿tica se manifiestan como aumentos en la temperatura, mientras que aumentos en la energ鱈a potencial se manifiestan como aumentos en el volumen. Tambi辿n describe los diferentes tipos de procesos termodin叩micos como adiab叩ticos, isob叩ricos e isot辿rmicos.
ACRL 2015 Presentation: Models of Library Engagement with MOOCsLaura O'Brien
油
Panel discussion description:
Discover how libraries at national and international higher education institutions are engaging with the design, production, and implementation of MOOCs at their campuses. What strategies have libraries employed to find the time, money, and staff necessary to develop models of support for production of (and teaching in) MOOCs, in addition to other distance learning initiatives? What you learn may influence whether and how deeply your library engages with MOOCs at your own institution.
1) La iluminaci坦n es la medida del flujo luminoso incidente sobre una superficie. Se define como el flujo luminoso por unidad de 叩rea y se mide en lux. 2) La intensidad luminosa de una fuente es el flujo luminoso emitido por unidad de 叩ngulo s坦lido y se mide en candelas. 3) La iluminaci坦n de una superficie depende de la intensidad de la fuente de luz y est叩 inversamente relacionada al cuadrado de la distancia entre la fuente y la superficie.
This poem explores themes of confusion, inner conflict, and the struggle to find clarity or meaning through references to sinking, hidden tricks, and paradoxes. The speaker seems lost in their thoughts and unable to break free from an endless cycle of questions without answers that leave them feeling perplexed and confined in their own mind.
El presente proyecto tuvo como objetivo evaluar un Manual de Funciones Administrativas dirigido al Consejo Comunal Sector 4 Urbanizaci坦n La Carucie単a. Se realiz坦 mediante una investigaci坦n de campo descriptiva con una muestra de 51 voceros y voceras del Consejo Comunal, a quienes se les aplicaron un cuestionario y una encuesta para indagar sobre la aplicaci坦n de las funciones establecidas en el Manual. Los resultados mostraron que si bien los voceros conocen el Manual, su implementaci坦n es escasa, pues se observ坦 poca informaci坦n para ofrec
O documento discute como fraudes podem ocorrer tanto no setor p炭blico quanto no privado, contrariando a afirma巽達o de que "Fraudes s坦 ocorrem no setor p炭blico!". Ele fornece estat鱈sticas mostrando que 5% do PIB mundial 辿 perdido para fraudes e estima que no Brasil isto representaria R$207 bilh探es perdidos. Alerta as empresas que 辿 preciso estar preparado contra fraudes pois a oportunidade e a falta de preparo podem levar a sua ocorr棚ncia.
Este documento presenta conceptos fundamentales de luminotecnia, incluyendo las magnitudes y unidades utilizadas como flujo luminoso, intensidad, iluminaci坦n y brillo. Explica la ley de Weber-Fechner que rige la relaci坦n no lineal entre est鱈mulo y sensaci坦n para la vista y otros sentidos. Tambi辿n describe las caracter鱈sticas y tipos de fuentes de luz, y c坦mo var鱈an la intensidad y iluminaci坦n con la distancia y 叩ngulo de incidencia de los rayos de luz.
During the planning and research stage for her project, Ifrah Khalif used various media technologies. She conducted online research using YouTube to watch previous student work and film trailers in the horror genre. Google was also used for research of film posters and magazines. Her mobile phone helped with organization and communication. A blogger was created to document the process and share research sources like questionnaires and schedules. Adobe software like Photoshop, After Effects, and Premiere Pro were used for editing tasks like the film poster, production ident, and final video.
The document contains two stories. The first story is about a 100 year old tree on a university campus that has witnessed history, including soldiers assembling under it before leaving for World Wars 1 and 2. Experts say the tree is remarkably healthy for its age and should live another 50-75 years. The second story is about an experienced hiker, Will Henderson, who broke his leg after slipping in a stream off the Appalachian Trail. Despite being experienced, he hiked alone and got lost. He crawled for 4 days with a makeshift splint before being rescued. He plans to finish hiking the trail but will not hike alone off the main path again.
An overview of 'Step by Step', the final report for the Shared Decision Making in BC.
The Shared Decision Making (SDM) in British Columbia Project was a multi-year independent research initiative housed at Simon Fraser Universitys Centre for Dialogue. This collaborative project examined the emergence of non-treaty SDM Agreements, negotiated between British Columbia and individual or groups of First Nations, in the form of Strategic Engagement Agreements or Reconciliation Protocols. Each SDM Agreement provides a framework for collaboration between two governments who despite their differences, seek to build working relationships with one another, develop trust, and find ways to reach mutually agreeable decisions about how land and resources should be managed.
The overarching purpose of the SDM in BC project was to understand more clearly where SDM agreements have come from, what they mean, and how they are working so far.
1. ANALIZA UTICAJA GLOBALNOG OGRANIENJA CE2-1
NA PREPOZNAVANJE IZOLOVANO IZGOVORENIH REI
SRPSKOG JEZIKA
Branko Markovi1
, Mirjana Miti1
, Milan Pajki1
, Nenad Milenkovi1
, Gordana Markovi2
REZIME
U ovom radu obja邸njeno je kako kori邸enje globalnog ogranienja CE2-1 utie na
prepoznavanje izolovano izgovorenih rei. Renik koji je kori邸en sastoji se od 邸est
osnovnih boja i etrnaest brojeva srpskog jezika. Algoritam za prepoznavanje je DTW.
Kljune rei: DTW algoritam, globalno ogranienje, kepstralni, delta kepstralni,
delta-delta kepstralni koeficijenti, prepoznavanje govora.
THE ANALYSIS OF THE GLOBAL CONSTRAIN CE2-1
INFLUENCE ON SERBIAN ISOLATED WORDS RECOGNITION
ABSTRACT
In this paper we explained how usage of the global constrain CE2-1 influence to
the recognize rate of isolated Serbian words. For the vocabulary we used six basic colors
and fourteen numbers. The algorithm for recognition was DTW (Dynamic Time Warping).
Key words: DTW algorithm, global constrains, cepstral, delta cepstral, delta-delta
cepstral coefficients, speech recognition.
1. UVOD
Pri re邸avanju problema prepoznavanja govora koriste se razliite metode i
razliiti algoritmi. Jedan od naje邸e kori邸enih metoda je DTW (Dynamic Time
Warping). Ovaj metod je 邸iroko primenjivan od samih poetaka prepoznavanja
govora i modifikovanjem razliitih parametara mogue je dobiti veu ili manju
uspe邸nost prepoznavanja.
U ovom radu glavni fokus je usmeren na razmatranje kako ogranienje
globalne staze utie na prepoznavanje govora. Razmatrana su dva osnovna sluaja:
1
Visoka 邸kola tehnikih strukovnih studija aak
2
Tehnika 邸kola, aak
2. kada ne postoji ogranienje i kada je ogranienje tipa CE2-1 (constranins
endpoints) [1]. Za bazu podataka kori邸en je skup uzoraka koji je snimljen na
Visokoj 邸koli tehnikih strukovnih studija aak poznat pod nazivoma Whi-Spe
[2]. U snimanju baze uestovalo je 10 studenta volontera koji su renik od 50
razliitih rei proitali po 20 puta (10 normalnim govorom i 10 邸apatom). Za
potrebe istra転ivanja koje je prikazano u ovom radu kori邸eni su izgovori dobijeni
pri normalnom govoru i pri tome samo deo baze koji se odnosi na boje i brojeve
srpskog jezika.
Cilj ovog rada je da se na bazi dobijenih rezultata utvrdi uticaj globalnog
ogranienja CE2-1 na verovatnou prepoznavanja govora za razliita vektorska
obele転ja.
2. DTW ALGORITAM
Govorni uzorci za ovo istra転ivanje snimani su na frekvenciji odmeravanja
od 22,050Hz, a potom sme邸tani u obliku wave fajlova u bazu podataka. Za jednu
re postoji po 10 govornih uzoraka normalnog govora i prilikom poreenja prvi
uzorak je kori邸en kao referentni, a ostalih devet su kori邸eni za poreenje. U
tabelama 1 i 2 date su rei koji su kori邸ene za ove eksprimente.
Tabela 1: Lista boja
BOJE
bela 転uta crna crvena plava zelena
Tabela 2: Lista brojeva
BROJEVI
nula jedan dva tri etiri pet 邸est
sedam osam devet deset sto hiljadu milion
Svaki od wave fajlova je dovoen na sistem za predobradu [3] koji je
prikazan na slici 1. Na osnovu ove predobrade dobijani su odgovarajui vektori: ili
kepstralnih koeficijenata, ili kepstralnih i delta kepstralnih ili kepstralnih i delta i
delta-delta kepstralnih koeficijenata [4].
Slika 1: Predobrada govornog signala
Vektori od kepstralnih koeficijenata sadr転e po 12 elemenata. Vektori od
kepstralnih i delta kepstralnih koeficijenata sadr転e 24 elementa od kojih su prvih 12
kepstralni, a sledeih 12 delta kepstralni koeficijenti. Vektori od kepstralnih, delta
3. kepstralnih i delta-delta kepstralnih koeficijenata sadr転e 36 elemenata od kojih su
prvih 12 kepstralni koeficijenti, sledeih 12 su delta kepstralni koeficijenti i zadnjih
12 su delta-delta kepstralni keoficijenti.
DTW algoritam omoguava pronala転enje optimalne staze izmeu poetnih
i krajnjih taaka kroz itavu mre転u potencijalnih prelaza [5]. Ako je poetna taka
(1,1), a krajnja (M,N), (gde je za primer sa slike 2 M=10, a N=12), onda ima vi邸e
razliitih mogunosti da se od poetne doe do krajnje take. Pri tome se rauna
cena ko邸tanja svakog od lokalnih prelaza.
Slika 2: Izbor optimalne staze
Pri ovim prelazima mo転e se koristi i razliit tip lokalnog ogranienja. Za
sluaj ovog istra転ivanja kori邸eno je lokalno ogranienje Tipa I [6].
DTW algoritam je rekurzivan postupak i mo転e se opisati kroz sledee
korake [5]:
1) Inicijalizacija:
)1(*)1,1()1,1( mdD =
(1)
2) Rekurzija:
[ ])),(),','(()','(min),(
','
jijidjiDjiD
ji
+=
(2)
3) Zavr邸etak:
陸
=
M
NMD
NMd
),(
),(
(3)
gde je d(i,j) lokalna distanca izmeu vektora i i j, D(i,j) je akumulirana distanca za
4. globalnu stazu do take (i,j), a 陸M je normalizacioni faktor.
Vrlo je va転no da se uva informacija o pre転ivelim stazama od take (1,1)
do take (M,N), a zatim idui unazad od take (M,N),da se rekonstrui邸e optimalna
staza [7].
3. GLOBALNA OGRANIENJA
Razvijeni softverski paket WiseWave 1.5 je baziran na DTW-u i
omoguava da se za potrebe ovog eksperimenta koriste razliita vektorska obele転ja
(kepstralni, delta kepstralni i delta-delta kepstralni koeficijenti), a takoe i da se
ukljui globalno ogranienje tipa CE2-1 ili da se testiranje vr邸i bez ogranienja
globalne staze.
Na slici 3 dat je prikaz zasenenog prostora po kome se vr邸i tra転enje
optimalne staze za sluaj a) bez globalnog ogranienja i sluaj b) kada je
ogranienje tipa CE2-1.
Slika 3: Bez globalnog ogranienja (a) i sa ogranienjem tipa CE2-1(b)
Na bazi ovog softvera firmirano je 邸est razliitih scenarija:
1) Vektor od kepstralnih koeficijenata (ukupno 12) i globalna staza bez
ogranienja;
2) Vektor od kepstralnih i delta kepstralnih koeficijenata (ukupno 24) i
globalna staza bez ogranienja;
3) Vektor od kepstralnih, delta kepstralnih koeficijenata i delta-delta
kepstralnih koeficijenta (ukupno 36) i globalna staza bez ogranienja;
4) Vektor od kepstralnih koeficijenata (ukupno 12) i globalna staza sa
CE2-1 ogranienjem;
5) Vektor od kepstralnih i delta kepstralnih koeficijenata (ukupno 24) i
globalna staza sa CE2-1 ogranienjem;
5. 6) Vektor od kepstralnih, delta kepstralnih koeficijenata i delta-delta
kepstralnih koeficijenta (ukupno 36) i globalna staza sa CE2-1
ogranienjem.
Detaljno su razmotrene ove razliite kombinacije i odgovarajui rezultati
dati u obliku tabela.
4. REZULTATI
Na bazi kreiranog softvera i kori邸enjem svih deset govornika dobijeni su
pojedinani rezultati. Prvih pet govornika su bile 転enske, a drugih pet mu邸ke osobe.
Rezultati su klasifikovani na one koji su dobijeni bez globalnog
ogranienja i na one sa CE2-1 globalnim ogranienjem.
4.1 Bez globalnog ogranienja
U tabelama 3 i 4 dati su sumarno rezultati za svakog od govornika kada
nema globalnog ogranienja (prema slici 3a).
Tabela 3: Uspe邸nost prepoznavanja govora za boje bez ogranienja (u %)
Vrsta
obele転ja/Govorni
k
Kepstralni (12)
Delta i Kepstralni
(24)
Delta-Delta i
Delta i Kepstralni
(36)
Govornik 1 100 100 100
Govornik 2 100 100 100
Govornik 3 96,29 96,29 96,29
Govornik 4 100 100 100
Govornik 5 96,29 98,15 98,15
Govornik 6 94,44 94,44 94,44
Govornik 7 92,59 90,74 90,74
Govornik 8 100 100 100
Govornik 9 96,29 96,29 96,29
Govornik 10 100 96,29 96,29
Tabela 4: Uspe邸nost prepoznavanja govora za brojeve bez ogranienja (u %)
Vrsta
obele転ja/Govorni
k
Kepstralni (12)
Delta i Kepstralni
(24)
Delta-Delta i
Delta i Kepstralni
(36)
Govornik 1 100 100 100
Govornik 2 100 100 100
Govornik 3 100 100 100
Govornik 4 100 100 100
Govornik 5 100 100 100
6. Govornik 6 100 100 100
Govornik 7 100 100 100
Govornik 8 100 100 100
Govornik 9 100 100 100
Govornik 10 100 100 100
4.2 Sa CE2-1 globalnim ogranienjem
U tabelama 5 i 6 prikazani su rezultati za sluaj kori邸enja CE2-1
globalnog ogranienja (Slika 3b).
Tabela 5: Uspe邸nost prepoznavanja govora za boje sa globalnim ogranienjem
CE2-1 (u %)
Vrsta
obele転ja/Govorni
k
Kepstralni
Delta i Kepstralni
(24)
Delta-Delta i
Delta i Kepstralni
(36)
Govornik 1 100 100 100
Govornik 2 100 100 100
Govornik 3 96,29 96,29 96,29
Govornik 4 100 100 100
Govornik 5 98,15 100 96,29
Govornik 6 94,44 94,44 94,44
Govornik 7 90,74 90,74 90,74
Govornik 8 100 100 100
Govornik 9 96,29 96,29 96,29
Govornik 10 100 96,29 96,29
Tabela 6: Uspe邸nost prepoznavanja govora za brojeve sa globalnim ogranienjem
CE2-1 (u %)
Vrsta
obele転ja/Govorni
k
Kepstralni
Delta i Kepstralni
(24)
Delta-Delta i
Delta i Kepstralni
(36)
Govornik 1 100 100 100
Govornik 2 99,21 99,21 99,21
Govornik 3 100 100 100
Govornik 4 100 100 100
Govornik 5 100 100 100
Govornik 6 100 100 99,21
Govornik 7 100 100 100
Govornik 8 100 100 100
Govornik 9 100 100 100
Govornik 10 100 100 100
7. Analizirajui dobijene rezultate za 10 raspolo転ivih govornika mo転e se
uoiti da je verovatnoa prepoznavanja za brojeve ne邸to vea nego za boje.
5. ZAKLJUAK
Na bazi izvr邸enih testiranja uoava se sledee:
prosena uspe邸nost prepoznavanja za boje kada nema globalnog
ogranienja je: 97,34%
prosena uspe邸nost prepoznavanja za brojeve kada nema globalnog
ogranienja je: 100%
prosena uspe邸nost prepoznavanja za boje kada je CE2-1 globalno
ogranienje primenjeno iznosi: 97,34%
prosena uspe邸nost prepoznavanja za brojeve kada je CE2-1 globalno
ogranienje primenjeno iznosi: 99,89%
Na osnovu ovih rezultata uoava se da je uspe邸nost prepoznavanja za boje i
brojeve pojedinano gotovo identina i za sluaj kada nema i kada ima globalnog
ogranienja. Takoe mo転e se zakljuiti da uticaj vrste vektora (kepstralni, delta i
delta-delta) nema znaajniju ulogu. Ovo se mo転e objasniti time 邸to su uzorci
snimani u uslovima potisnutog ambijentalnog 邸uma, pa kako delta i delta-delta
utiu na potiskivanje 邸uma to njihov znaaj nije do邸ao do izra転aja.
Mo転e se uoiti da je verovatnoa prepoznavanja brojeva za oko 2,6% bolja
nego verovatnoa prepoznavanja boja. To se mo転e objasniti injenicom da su
pojedine boje fonetski dosta sline i naje邸a konfuzija je upravo izmeu samo par
boja.
Kao generalan zakljuak mo転e se konstatovati da za kori邸enu bazu
podataka Whi-Spe primenjeni DTW postupak daje odline rezultate u
prepoznavanju normalnog govora. Pretpostavka da e uvoenje ogranienja tipa
CE2-1 dati bolje rezultate nije se ostvarila jer i pri scenariju bez ogranienja
prepoznavanje je vrlo visoko. Dalja istra転ivanja mogu biti usmerena ka
prepoznavanju 邸apata kao i kori邸enju drugih vrsta obele転ja i ogranienja (lokalnog
i globalnog tipa).
6. LITERATURA
[1] L.R. Rabiner, S.E. Levinson, A.E. Rosenberg, J.G. Wilpon,
Speaker-Independent Recognation of Isolated Words Using
Clustering Techniques, IEE Transaction on Acoustics, Speech adn
Signal Procesing, Vol ASSP-27, No.4, 1979. pp. 336-349.
8. [2] B. Markovi, S. T. Jovii, J. Gali, . Grozdi, Whispered
speech database: design, processing and application, The 16th
International Conference TSD 2013, Pilsen, Czech Republic, 2013,
pp. 591-598.
[3] G. Markovi, B. Markovi: Vizuelni DTW kao nastavno sredstvo
za poreenje govornih uzoraka, TIO 2008, pp. 409-415, 2008.
[4] B. Markovi, J. Gali, . Grozdi, S. T. Jovii, Application of
DTW method for Whispered Speech Recognition, The 4th
International Conference on Fundamental and Applied Aspects of
Speech and Language, Belgrade, Serbia, October 25-26.
[5] L. Rabiner, B-H. Juang, Fundamentals of Speech Recognition,
Prentice Hall, 1993.
[6] H. Sakoe, S. Chiba, Dynamic programming optimization for
spoken word recognition, IEEE Trans. Acoustics, Speech, Signal
Proc., ASSP-26(1):43-49, February 1978.
[7] R. Mitrovi, I. 貼ivanovi, I. Radelji, B. Markovi, Kori邸enje
asinhronog dinamikog programiranja u re邸avanju problema
optimalne staze, TIO 2012, aak, 2012.