Management and Future Technologies | Volume 1, Issue 1 | № 001

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Asosiy qism Taklif etililayotgan ishda yuzni tanib olish, boshning, ko‘zning holatinining muvofiqligi tekshirish hamda kerakli karorlarni qabul qilishdan tashqari ushbu tizimdan foydalanish jarayonida yaniy imtihon vaqtidan bir necha yuzlar, ko‘shimcha tovushlar va aksessuarlar aniqlab xulosalar taqdim etiladi. Masofaviy ta’lim jarayonida videomonitoring tizimini ishlab chiqishda quyidagi bosqichlarni o‘zi mujassamlagan bo‘lishi kerak. Foydalanuvchi bosqichi; Hizmat ko‘rsatuvchi texnik bosqich; Yakuniy xuloslarni taqdim etish bosqichi; Yuqorida taklif etilgan ko‘p bosqichli videomonitoring tizimning ichki modullari mavjud. Ma’lumotlarni to‘plash va saqlash moduli – Ushbu modulda talaba kompyuterining veb kamerasi, mikrafoni va klavyaturasidan malumotlar olinadi. Ushbu malumotlar foydalanuvchi identifikatsiya kodi bilan serverga shifrlangan SSL kanali orqali yuboriladi. Autentifikatsiya moduli - Barcha ma’lumotlar serverlarda saqlanganidan so‘ng, talaba ro‘yxatdan o‘tish vaqtida yaratilgan va oxirgi ijobiy ma’lumotlar bilan yangilangan identifikatsiya kodi bilan bog‘langan biometrik model bilan taqqoslanadi. Natija tizimning ma’lumotlar bazasida saqlanadi. Proktoring (nazorat) va kompyuter bloklash moduli - Videomonitoring tizimi onlayn mashg‘ulotlar davomida yozib oling tasvir va audio yozuvlar yozib boriladi va mashg‘ulotlar davomida talabalar nomaqbul xatti-harakatlarini aniqlash uchun kayta ishlov beriladi. Shundan so‘ng tizim talaba mashg‘ulot va nazoratlarni topshirish vaqtida ko‘shimcha yordam olayotganligini (telfon orqali, yonidagi boshqa talaba) yoki taqiqlangan hujjatlardan (elektron kitoblar, maruza mantlar va internetga ulangan boshqa qurilmalar....) foydalanmasligi nazorat qila oladi. Bundan tashqari, agar biror talaba tizimni aldashga harakat qilsa, masalan, kamera oldiga fotosurat o‘rnatish yoki ID karta tasvirini boshqa birovning kartasiga almashtirish kabi aldashlarga urinsa tizim tomonidan aniqlanadi va nazoratchi xabar qilinadi. Insonni tekshirish moduli: Bunda registrasiya qilingan malumotlar bilan sessiya davomida olingan ovozli yozuvlar, videotasvirlar va tugmachalarni bosish namunalarini tasodifiy malumotlari bilan (turli odamlar tomonidan tahlil qilingan bir xil videotasvirlar, ovozili yozuvlar va tugmachalarni bosish namunalari) o‘zaro tekshiruviga asoslanadi. Bundan tashqari, fotosuratlar yoki ovoz sifati yaxshi bo‘lmagan taqdirda, maxsus o‘qitilgan xodimlar talabaning 100% ishonch tekshiruvini ta’minlaydigan shaxs ishtokidagi yuzma yuz tekshiruvi o‘tkaziladi. Natijalarni ko‘rish moduli: Yakuniy nazorat natijalari malumotlarni ushbu moduli tomonidan taqdim etiladi. U dinamik veb-sahifa sifatida 24 soat/365 kun talab bo‘yicha grafik diagrammalar va jadvallarni yaratadi. Yakuniy hisobotlarni turli formatlarda yuklab olish yoki chop etish mumkin. Bundan tashqari, malumotlarni taqdim etish moduli oldindan belgilangan taqiqlangan xatti-harakatlar sodir bo‘lganda avtomatlashtirilgan signallarni ishlab chiqaradi. Tizim autentifikatsiya jarayonlari fakat foydalanuvchilar tizimga kirganlarida emas, balki butun sesciya davomida baholashdan tashqari fan bo‘yicha shakillantirilagan maruza va amaliyot materiallarini o‘zlashtirish jarayonlarini ham nazorat qilishi lozim. 21

Volume 1, Issue 1 22 April 2024 Management and Future Technologies journal.umft.uz Adabiyotlar 1. Arora, P. (2021). Is Remote Proctoring The Future Of Academia? - eLearning Industry. 2. Atoum, Y., Chen, L., Liu, A. X., Hsu, S. D. H., & Liu, X. (2017). Automated Online Exam Proctoring.IEEE Transactions on Multimedia, 19(7), 1609–1624. https://doi.org/10.1109/ TMM.2017.2656064 Beust, P., Duchatelle, I., & Cauchard, V. 3. S.Xalilov, M.Mardiyev “Masofaviy ta’lim jarayonida avtoproktoring tizimlarining istiqbollari” juornal Innovations in technology and science education SJIF 2023: 5.503 volume 2 ISSUE 8. 4. Draaijer, S., Jefferies, A., & Somers, G. (2018). Online proctoring for remote examination: A state of play in higher education in the EU. Communications in Computer and Information Science, 829, 96–108. https://doi.org/10.1007/978-3-319-97807-9_8. 5. Caveon, D. F., Security, T., & Layman, H. (2013). Online Proctoring Systems Compared. 6. Chua, S. S., Bondad, J. B., Lumapas, Z. R., & Garcia, J. D. (2019). Online Examination System with Cheating Prevention Using Question Bank Randomization and Tab Locking. 7. Coghlan, S., Miller, T., & Paterson, J. (2020). Good proctor or “Big Brother”? AI Ethics and Online Exam Supervision Technologies. ArXiv Preprint. 8. Dendir, S., & Maxwell, R. S. (2020). Cheating in online courses: Evidence from online proctoring. Com- puters in Human Behavior Reports, 2, 100033. https://doi.org/10.1016/j.chbr.2020.100033

Management and Future Technologies journal.umft.uz ELEKTRON SO‘ZLARNI VEKTORLI IFODALASH TEXNOLOGIYASI A nnotatsiya: Tabiiy tilni qayta ishlash texnologiyalarini ishlab chiqish va samarali foydalanishni tashkil etish bugungi kunning dolzarb maslalaridan biridir. Ushbu ishda elektron matndagi so‘zlarni qidirish va ma’no jihatidan tahlil qilishni tashkil etishda qo‘llaniladigan samarali texnologiyalardan biri bo‘lgan so‘zni vektorli ifodalash texnologiyalari haqida so‘z yuritilgan. Xususan, Word2vec texnologiyasi bo‘yicha so‘zni vektor qiymatlarini aniqlash jarayoni tushuntirilgan. April 2024 Volume 1, Issue 1 23 Muallif Baxtiyor Akmuradov Kommunikatsiya va raqamli texnologiyalar kafedrasi, “University of Management and Future Technologies” universiteti, Toshkent, 100208, O‘zbekiston; b.u.akmuradov@gmail.com Mas’ul: b.u.akmuradov@gmail.com Kalit so‘zlar: matn, so‘z, texnologiya, vektor, Word2vec, Ochiai koeffitsiyenti, GloVe. Copyright: © 2024 by the authors. Ushbu maqola Creative Commons Attribution (CC BY) litsenziyasi shartlari asosida tarqatiladigan ochiq foydalanish maqolasi hisoblanadi (https://creativecommons.org/ licenses/by/4.0/).

Volume 1, Issue 1 April 2024 24 Authors Bakhtiyar Akmuradov Department of Communication and Digital Technologies, University of Management and Future Technologies, Tashkent, 100208, Uzbekistan; b.u.akmuradov@gmail. com Management and Future Technologies journal.umft.uz TECHNOLOGY OF VECTOR REPRESENTATION OF ELECTRONIC WORDS A bstract: The development and effective use of medical language processing technologies is one of today’s urgent issues. This paper talks about word vector representation technologies, which are one of the effective technologies used in the organization of searching and semantic analysis of words in electronic text. In particular, the process of determining word vector values according to Word2vec technology is explained. Keywords: text, word, technology, vector, Word2vec, Ochiai coefficient, GloVe. Copyright: © 2024 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/ by/4.0/).

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Kirish Hisoblash mashinalari yordamida matnni elektron tarzda saqlash, o‘zgartirish va uzatishni amalga oshiruvchi ko‘plab vositalar mavjud. Zamonaviy sun’iy intellekt texnologiyalarining, NLP (Natural Language Processing) texnologiyalarining rivojlanishi bilan tabiiy tilga ishlov berish vositalari soni va sifati ham keskin o‘sdi[1]. Bunda matn tahlili, so‘zlarni tanib olish, guruhlash va ma’no jihatidan tahlil qilish hamda ranjirlashni amalga oshirish uchun so‘zni kompyuter tushunadigan tilga o‘girish talab etiladi[3]. So‘zning barcha xususiyatlarini inobatga olgan holda raqamli ko‘rinishning samarali usullaridan biri bu vektorli ifodalashdir. So‘zlarni vektorli ifodalashni hozirda keng qo‘llanilayotgan texnologiyalardan biri misolida tushuntirishga harakat qilamiz. Word2vec – so‘zlarni vektorli ifodalashning samarali texnologiyalaridan biri bo‘lib, u 2013-yilda ishlab chiqilgan. So‘zlar bilan ishlashdan tashqari, uning ba'zi kontsepsiyalari tavsiya berish mexanizmlarini ishlab chiqishda va ma'lumotlar ma'nosini ifodalashda, hatto tijoriy, lingvistik bo‘lmagan vazifalarda ham samarali qo‘llanilib kelinmoqda[6]. Xusususan, Airbnb, Alibaba, Spotify va Anghami kabi kompaniyalar ushbu texnologiyadan o‘zlarining tavsiya tizimlarida foydalanganlar[2]. Asosiy qism misolni ko‘rib chiqamiz. Masalan: bir insonni tasvirlash uchun uning bir nechta Ob'ektlarni vektor ko‘rinishida tasvirlashni sodda tilde tushuntirish uchun quyidagi misolni ko‘rib \YWYWM]EXPEVMRME]XMWLQYQOMR<YWYWYWM]EXPEVWSRMRMRKSVXMFFSVMWLMFMPERXEWZMVPERE]SXKER misolni chiqamiz. insonni tasvirlash uchun uning bir mumkin. nechta chiqamiz. ko‘rib Masalan: bir insonniMasalan: tasvirlash bir uchun uning bir nechta xususiyatlarini aytish shaxs haqida aniqroq tasovvur uyg‘ona boshlaydi. \YWYWM]EXPEVMRME]XMWLQYQOMR<YWYWYWM]EXPEVWSRMRMRKSVXMFFSVMWLMFMPERXEWZMVPERE]SXKER Xusususiyatlar sonining ortib borishi bilan tasvirlanayotgan shaxs haqida aniqroq tasovvur uyg‘ona Misol uchun biz tanlagan 0 dan 100 gacha bo‘lgan shkalada shaxs haqida aniqroq tasovvur shaxsni uyg‘ona tasvirlash boshlaydi.uchun boshlaydi. 5 ta xususiyatlarini baholash kerak bo‘lsin: Misol uchun biz tanlagan shaxsni tasvirlash uchun 0 dan 100 gacha bo‘lgan shkalada Misol uchun biz tanlagan shaxsni tasvirlash uchun 0 dan 100 gacha bo‘lgan shkalada 5 ta 5 ta xususiyatlarini baholash kerak bo‘lsin: xususiyatlarini baholash kerak bo‘lsin: NEHZEP<YWYWM]EXPEVNEHZEPM NEHZEP<YWYWM]EXPEVNEHZEPM &ELS № <YWYWM]EX HEKEGLESVEPMUHE &ELS № <YWYWM]EX ,MWWM]SXKEFIVMPYZGLERPMO HEKEGLESVEPMUHE /YGLPMOYGLWM^ ,MWWM]SXKEFIVMPYZGLERPMO Sog‘lom/xasta /YGLPMOYGLWM^ =SWLUEVM Sog‘lom/xasta &EPERHTEWX =SWLUEVM &EPERHTEWX Jadvaldan ko‘rinib turibtiki tasvirlanayotgan shaxsning hissiyotga beriluvchanlik 1-jadval. Xususiyatlar jadvali HEVENEWM KE XIRK ZE LO ,MWSFPEWL QEWLMREWMHE EQEPKE SWLMVMPEHMKER OMXSFPEVRM Jadvaldan ko‘rinib turibtiki tasvirlanayotgan shaxsning hissiyotgaLMWSF beriluvchanlik WSHHEPEWLXMVMWL YGLYR UM]QEXPEVRM 1 da +1 oralig‘idagi shkalaga o‘tkazsak quyidagi HEVENEWM KE XIRK ZE LO ,MWSFPEWL QEWLMREWMHE EQEPKE SWLMVMPEHMKER LMWSF OMXSFPEVRM Jadvaldan ko‘rinib turibtiki tasvirlanayotgan shaxsning hissiyotga beriluvchanlik darajasi 38 ga teng qiymatlarga ega bo‘lamiz: WSHHEPEWLXMVMWL YGLYR UM]QEXPEVRM 1 da +1 oralig‘idagi shkalaga o‘tkazsak quyidagi va hk. Hisoblash mashinasida amalga oshiriladigan hisob kitoblarni soddalashtirish uchun qiymatlarni qiymatlarga ega bo‘lamiz: -1 da +1 oralig‘idagi shkalaga o‘tkazsak quyidagi qiymatlarga ega bo‘lamiz: 7LE\W 7LE\W Tanlangan shaxga o‘xshash bo‘lgan shaxsni topish uchun yana 2 XE WLE\WRMRK turdagi xusususiyatlari bo‘lsin. Tanlangan shaxgaberilgan o‘xshash bo‘lgan shaxsni topish uchun yana 2 XE WLE\WRMRK turdagi xusususiyatlari berilgan bo‘lsin. 7LE\W 7LE\W 7LE\W 7LE\W Berilgan xususiyatlarga ega ikkita shaxsning qaysi biri tanlangan shaxsga ko‘proq o‘xshashligini uchun ega ularni vektor ko‘rinishida va taqqoslash Berilgan aniqlash xususiyatlarga ikkita shaxsning qaysiifodalash biri tanlangan shaxsgamumkin. ko‘proq 5Y]MHEKMVEWQHEYPEVRMRKZIOXSVMJSHEWMOIPXMVMPKER 25

26 WSHHEPEWLXMVMWL YGLYR UM]QEXPEVRM 1 da +1 oralig‘idagi shkalaga o‘tkazsak quyidagi qiymatlarga ega bo‘lamiz: Jadvaldan ko‘rinib turibtiki tasvirlanayotgan shaxsning hissiyotga beriluvchanlik KE XIRK ZE LO ,MWSFPEWL QEWLMREWMHE EQEPKE SWLMVMPEHMKER LMWSF OMXSFPEVRM HEVENEWM WSHHEPEWLXMVMWL YGLYRApril UM]QEXPEVRM 1 da Management +1 oralig‘idagi shkalaga o‘tkazsak quyidagi Volume 1, Issue 1 2024 and Future Technologies journal.umft.uz qiymatlarga ega bo‘lamiz: 7LE\W Tanlangan shaxga o‘xshash bo‘lgan shaxsni topish uchun yana 2 tauchun shaxsning 2 turdagi Tanlangan shaxga o‘xshash bo‘lgan shaxsni topish yana 2 XExusususiyatlari WLE\WRMRK berilgan bo‘lsin. 7LE\W turdagi xusususiyatlari berilgan bo‘lsin. Tanlangan shaxga7LE\W o‘xshash bo‘lgan shaxsni topish uchun yana 2 XE WLE\WRMRK turdagi xusususiyatlari berilgan bo‘lsin. 7LE\W 7LE\W Berilgan xususiyatlarga ega ikkita shaxsning qaysi biri tanlangan shaxsga ko‘proq Berilgan xususiyatlarga ega ikkita shaxsning qaysi biri tanlangan shaxsga ko‘proq o‘xshashligini 7LE\W ko‘rinishida o‘xshashligini aniqlash uchun ularni vektor ifodalash va taqqoslash mumkin. aniqlash uchun ularni vektor ko‘rinishida ifodalash va taqqoslash mumkin. Quyidagi 1-rasmda 5Y]MHEKMVEWQHEYPEVRMRKZIOXSVMJSHEWMOIPXMVMPKER ularning vektor ifodasi keltirilgan. Berilgan xususiyatlarga ega ikkita shaxsning qaysi biri tanlangan shaxsga ko‘proq o‘xshashligini aniqlash uchun ularni vektor ko‘rinishida ifodalash va taqqoslash mumkin. 1-rasm. Shaxslar xususiyatlarining vektorli ifodasi 5Y]MHEKMVEWQHEYPEVRMRKZIOXSVMJSHEWMOIPXMVMPKER VEWQ7LE\WPEV\YWYWM]EXPEVMRMRKZIOXSVPMMJSHEWM Keltirilgan vektorli ifodadan qaysi bir shaxs tanlangan shaxsga ko‘proq o‘xshashligi haqida fikr yuritish mumkin. Biroq, hisoblash mashinalarida vektorlar orasidagi o‘xshashlik VEWQ7LE\WPEV\YWYWM]EXPEVMRMRKZIOXSVPMMJSHEWM SHEXHE 3GLMEM OSIJJMXWMIRXM KISQIXVMO OSIJJMXWMIRX FMPER ERMUPEREHM &IVMPKER MOOM Keltirilgan vektorli ifodadan qaysi bir shaxs tanlangan shaxsga ko‘proq o‘xshashligi haqida fikr Keltirilgan bir shaxs tanlangan shaxsga ko‘proqodatda o‘xshashligi yuritish mumkin.vektorli Biroq, ifodadan hisoblash qaysi mashinalarida vektorlar orasidagi o‘xshashlik Ochiai haqida fikr yuritish mumkin. Biroq, hisoblash mashinalarida vektorlar orasidagi o‘xshashlik koeffitsienti (geometrik aniqlanadi. Berilgan ikki shaxsning tanlanganbo‘yicha shaxsga shaxsning tanlangan koeffitsient) shaxsga bilan o‘xshashlik darajasini Ochiai koeffitsienti SHEXHE 3GLMEM OSIJJMXWMIRXM KISQIXVMO OSIJJMXWMIRX FMPER ERMUPEREHM &IVMPKER MOOM o‘xshashlik darajasini Ochiai koeffitsienti bo‘yicha hisoblasak quyidagi natijaga ega bo‘lamiz. hisoblasak quyidagi natijaga ega bo‘lamiz. shaxsning tanlangan shaxsga o‘xshashlik darajasini Ochiai koeffitsienti bo‘yicha hisoblasak quyidagi natijaga ega bo‘lamiz. O‘xshashlik ( ! O‘xshashlik ( ! ( O‘xshashlik ! O‘xshashlik ( ! ko‘rish mumkinki birinchi tanlangan ko‘proq o‘xshashlikka Natijadan ko‘rish mumkinki birinchi shaxsshaxs tanlangan shaxsgashaxsga ko‘proq o‘xshashlikka ega. Biroq, Natijadan ega. Biroq, ko‘rilgan misolda faqat 2 ta xususiyat bo‘yicha vektor qiymatlarining ko‘rilgan misolda faqat 2 ta xususiyat bo‘yicha vektor qiymatlarining o‘xshashligi hisoblandi. Shaxsning Natijadan ko‘rish mumkinki birinchi shaxs tanlangan shaxsga ko‘proq o‘xshashlikka o‘xshashligi hisoblandi. Shaxsning taqqoslash mumkin bo‘lgan TEVEQIXVPEVM WSRM SVXMF taqqoslash mumkin bo‘lgan parametrlari borishi bilanbo‘yicha vektor qiymatlarining o‘xshashligi ega. Biroq, ko‘rilgan misolda faqatsoni 2 ortib ta xususiyat vektor qiymatlarining borishi bilan vektor qiymatlarining o‘xshashligi yanada aniqroq ifodalash mumkinWSRM bo‘ladi. yanada aniqroq ifodalash mumkin bo‘ladi. o‘xshashligi hisoblandi. Shaxsning taqqoslash mumkin bo‘lgan TEVEQIXVPEVM SVXMF 3GLMEM OSIJJMXWM]IRXM ]ETSR FMSPSKM %OMVE 3GLMEM XSQSRMHER ]MPHE XEOPMJ UMPMRKER borishi bilan vektor qiymatlarining o‘xshashligi yanada aniqroq ifodalash mumkin bo‘ladi. Ochiai koeffitsiyenti yapon biologi Akira Ochiai tomonidan 1957 yildaumumlashtirilgan taklif qilingan ikkilik ikkilik o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida va 3GLMEM OSIJJMXWM]IRXM ]ETSR FMSPSKM %OMVE 3GLMEM XSQSRMHER ]MPHE XEOPMJ UMPMRKER o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida umumlashtirilgan va turli ilovalarda turli ilovalarda qo‘llanila boshlandi. ikkilik o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida umumlashtirilgan va qo‘llanila boshlandi. Standart ta'rifda ikkita ixtiyoriy A va B to‘plamlari uchun koeffitsiyent quyidagicha turli ilovalarda qo‘llanila boshlandi. Standart ta’rifda ikkitaikkita ixtiyoriy A va B Ato‘plamlari uchun koeffitsiyent quyidagicha hisoblanadi: LMWSFPEREHM Standart ta'rifda ixtiyoriy va B to‘plamlari uchun koeffitsiyent quyidagicha |A∩B| 𝐾𝐾 = LMWSFPEREHM √|𝐴𝐴|∙|𝐵𝐵| |A∩B| 𝐾𝐾 =(QSщRSWXM Ko‘p hollarda Ochiai o‘lchovi FY ]IVHE `%` A to‘plamning kardinalligi. √|𝐴𝐴|∙|𝐵𝐵| quyidagi shaklda`%` qo‘llaniladi: FY ]IVHE A to‘plamning (QSщRSWXM kardinalligi. Ko‘p hollarda Ochiai o‘lchovi |A∩B|2 Ta'riflovchi to‘plamlar uchun quyidagicha hisoblanadi: 𝐾𝐾 Ochiai = |𝐴𝐴|∙|𝐵𝐵|koeffitsienti quyidagi shaklda qo‘llaniladi: 𝐾𝐾 = |𝐴𝐴|∙|𝐵𝐵| |A∩B|2 ∑𝑟𝑟𝑖𝑖=1 min(𝐴𝐴𝑖𝑖 ,𝐵𝐵𝑖𝑖 ) 𝐾𝐾 Ochiai = quyidagicha Ta'riflovchi to‘plamlar uchun koeffitsienti hisoblanadi:

3GLMEM ]ETSR FMSPSKM %OMVE yanada 3GLMEM XSQSRMHER ]MPHE XEOPMJ UMPMRKER borishi bilan OSIJJMXWM]IRXM vektor qiymatlarining o‘xshashligi aniqroq ifodalash mumkin bo‘ladi. OSIJJMXWM]IRXM ]ETSR FMSPSKM %OMVE 3GLMEM XSQSRMHER umumlashtirilgan ]MPHE XEOPMJ UMPMRKER ikkilik3GLMEM o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida va 3GLMEM OSIJJMXWM]IRXM ]ETSR FMSPSKM %OMVE 3GLMEM XSQSRMHER ]MPHE XEOPMJ UMPMRKER ikkilik o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida umumlashtirilgan va turli ilovalarda qo‘llanila boshlandi. ikkilik o‘xshashlik o‘lchovi bo‘lib, keyinchalik biologiyadan tashqarida umumlashtirilgan va turli ilovalarda qo‘llanila boshlandi. Standart ta'rifda ikkita ixtiyoriy A va B to‘plamlari uchun koeffitsiyent quyidagicha turli ilovalarda qo‘llanila boshlandi. Management and ta'rifda Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Standart ikkita ixtiyoriy A va B to‘plamlari uchun koeffitsiyent quyidagicha LMWSFPEREHM Standart ta'rifda ikkita ixtiyoriy A va B to‘plamlari uchun koeffitsiyent quyidagicha LMWSFPEREHM |A∩B| 𝐾𝐾 = |A∩B| LMWSFPEREHM √|𝐴𝐴|∙|𝐵𝐵| 𝐾𝐾 = |A∩B| kardinalligi. Ko‘p hollarda Ochiai o‘lchovi quyidagi bu yerda, |A|- A to‘plamning (moщnosti) √|𝐴𝐴|∙|𝐵𝐵| 𝐾𝐾 =(QSщRSWXM Ko‘p hollarda Ochiai o‘lchovi FY ]IVHE `%` A to‘plamning kardinalligi. √|𝐴𝐴|∙|𝐵𝐵| shaklda qo‘llaniladi: FY ]IVHE `%` A to‘plamning (QSщRSWXM kardinalligi. Ko‘p hollarda Ochiai o‘lchovi quyidagi shaklda`%` qo‘llaniladi: FY ]IVHE A to‘plamning (QSщRSWXM kardinalligi. Ko‘p hollarda Ochiai o‘lchovi quyidagi shaklda qo‘llaniladi: |A∩B|2 quyidagi shaklda qo‘llaniladi: 𝐾𝐾 = |A∩B|2 |𝐴𝐴|∙|𝐵𝐵|2 𝐾𝐾 = |A∩B| 𝐾𝐾 Ochiai = |𝐴𝐴|∙|𝐵𝐵| Ta'riflovchi to‘plamlar uchun koeffitsienti quyidagicha hisoblanadi: |𝐴𝐴|∙|𝐵𝐵| Ta'riflovchi to‘plamlar uchun Ochiai quyidagicha hisoblanadi: Ta’riflovchi to‘plamlar uchun Ochiai koeffitsienti hisoblanadi: ∑𝑟𝑟𝑖𝑖=1koeffitsienti min(𝐴𝐴𝑖𝑖quyidagicha ,𝐵𝐵𝑖𝑖 ) Ta'riflovchi to‘plamlar uchun hisoblanadi: 𝐾𝐾 Ochiai = ∑𝑟𝑟𝑖𝑖=1koeffitsienti min(𝐴𝐴 ,𝐵𝐵𝑖𝑖 ) quyidagicha 𝑟𝑟 (𝐴𝐴 )∙∑𝑟𝑟 𝑖𝑖 (𝐵𝐵 𝐾𝐾 = √∑∑𝑟𝑟𝑖𝑖−1 ) min(𝐴𝐴 ,𝐵𝐵 ) 𝑖𝑖 𝑖𝑖 𝑖𝑖 𝑖𝑖 𝑖𝑖−1 𝑖𝑖=1 𝐾𝐾 = √∑𝑟𝑟𝑖𝑖−1(𝐴𝐴𝑖𝑖)∙∑𝑟𝑟𝑖𝑖−1(𝐵𝐵𝑖𝑖) √∑𝑟𝑟𝑖𝑖−1(𝐴𝐴𝑖𝑖 )∙∑𝑟𝑟𝑖𝑖−1(𝐵𝐵𝑖𝑖 ) Bunday o‘lchov ekologiyada ko‘plik namunalar uchun keng qo‘llaniladi. Agar ob'ektlar Bunday o‘lchov ekologiyada ko‘plik namunalar keng qo‘llaniladi. Agar ob'ektlar turlarning paydo bo‘lishi bilan solishtirilsa, ya'ni P uchun (A) ehtimolliklari XYVPEV YGLVE]HMKER Bunday o‘lchov ekologiyada ko‘plik namunalar uchun keng qo‘llaniladi. Agar ob'ektlar turlarning paydo bo‘lishi bilan solishtirilsa, ya'ni P (A) ehtimolliklari XYVPEV YGLVE]HMKER Bunday o‘lchov ekologiyada ko‘plik namunalar uchun keng qo‘llaniladi. Agar ob’ e ktlar turlarning REQYREPEVRMRK RMWFM] WSRM LMWSFKE SPMRWE Y LSPHE OSIJJMXWMIRX LSHMWEPEVRMRK QYZSJMUPMKM turlarning paydo bo‘lishi bilan solishtirilsa, ya'ni P (A) ehtimolliklari XYVPEV YGLVE]HMKER REQYREPEVRMRK RMWFM] WSRM LMWSFKE SPMRWE Y LSPHE-OSIJJMXWMIRX LSHMWEPEVRMRK QYZSJMUPMKM paydo bo‘lishi bilan solishtirilsa, ya’ni P (A) ehtimolliklari turlar uchraydigan namunalarning nisbiy bo‘yicha hisoblanadi: REQYREPEVRMRK RMWFM] WSRM LMWSFKEhodisalarning SPMRWE Y LSPHE OSIJJMXWMIRX LSHMWEPEVRMRK bo‘yicha hisoblanadi: soni hisobga olinsa, u holda koeffitsient muvofiqligi bo‘yicha hisoblanadi: QYZSJMUPMKM 𝑃𝑃(A∩B) bo‘yicha hisoblanadi: 𝐾𝐾0,0 = 𝑃𝑃(A∩B) 𝐾𝐾0,0 = √𝑃𝑃(𝐴𝐴)∙𝑃𝑃(𝐵𝐵) 𝑃𝑃(A∩B) 𝐾𝐾0,0 = √𝑃𝑃(𝐴𝐴)∙𝑃𝑃(𝐵𝐵) √𝑃𝑃(𝐴𝐴)∙𝑃𝑃(𝐵𝐵) &YRHERUY]MHEKM\YPSWEPEVRMGLMUEVMWLMQM^QYQOMR &YRHERUY]MHEKM\YPSWEPEVRMGLMUEVMWLMQM^QYQOMR 2EJEUEX SHEQPEV FEPOMchiqarishimiz FSWLUE SF]IOXPEVRM Bundan quyidagi xulosalarni mumkin: VEUEQPM ZIOXSVPEV WMJEXMHE MJSHEPEWL &YRHERUY]MHEKM\YPSWEPEVRMGLMUEVMWLMQM^QYQOMR 2EJEUEX SHEQPEV FEPOM FSWLUE SF]IOXPEVRM VEUEQPM mumkin. ZIOXSVPEV WMJEXMHE MJSHEPEWL mumkin. Vektorlar o‘rtasida o‘xshashlikni osongina hisoblash 2EJEUEX SHEQPEV FEPOM FSWLUE SF]IOXPEVRM VEUEQPM ZIOXSVPEV NafaqatVektorlar odamlar balki boshqa obyektlarni raqamli vektorlar sifatida ifodalashWMJEXMHE mumkin.MJSHEPEWL Vektorlar mumkin. o‘rtasida o‘xshashlikni osongina hisoblash mumkin. Ko‘rib chiqilgan xususiyatlarni so‘zlarga qo‘llash orqali ma’no jihatidan bir biriga yaqin mumkin. Vektorlar o‘rtasida o‘xshashlikni osongina hisoblash mumkin. o‘rtasida o‘xshashlikni osongina hisoblash mumkin. Ko‘rib chiqilgan so‘zlarga Buning qo‘llashuchun orqali ma’no jihatidan biriga yaqin so‘zlarni qidirish va xususiyatlarni ajratib olish mumkin. so‘zlarni ma’nobir jihatida to‘liq Ko‘rib chiqilgan xususiyatlarni so‘zlarga Buning qo‘llashuchun orqali ma’no jihatidan bir biriga yaqin so‘zlarni qidirish va ajratib olish mumkin. so‘zlarni ma’no jihatida to‘liq Ko‘rib chiqilgan xususiyatlarni so‘zlarga qo‘llash orqali ma’no jihatidan bir biriga yaqin so‘zlarni ifodalangan holdava vektor ko‘rinishiga o‘tkazish talab uchun etiladi. so‘zlarni Hozirgi kunda so‘zlarni vektor so‘zlarni qidirish ajratib olish mumkin. Buning ma’no jihatida to‘liq ifodalangan holda vektor o‘tkazish talab etiladi. Hozirgi so‘zlarni vektor qidirish va ajratib olish mumkin. Buning uchun so‘zlarni ma’no jihatida to‘liq kunda ifodalangan holdaUlarga vektor ko‘rinishida ifodalovchi birko‘rinishiga qator texnologiyalar va dasturlash kutubxonalari mavjud. ifodalangan holda vektor ko‘rinishiga o‘tkazish talab etiladi. Hozirgi kunda so‘zlarni vektor ko‘rinishida ifodalovchi qatorHozirgi texnologiyalar va dasturlash kutubxonalari mavjud.bir Ularga ko‘rinishiga o‘tkazish talabbir etiladi. kunda so‘zlarni vektor ko‘rinishida ifodalovchi qator UY]MHEKMPEVRMQMWSPUMPMWLQYQOMR ko‘rinishida ifodalovchi bir qator texnologiyalar va dasturlash kutubxonalari mavjud. Ularga UY]MHEKMPEVRMQMWSPUMPMWLQYQOMR texnologiyalar va dasturlash kutubxonalari mavjud. Ularga quyidagilarni misol qilish mumkin: ;SVHZIG UY]MHEKMPEVRMQMWSPUMPMWLQYQOMR -;SVHZIG +PS:I ;SVHZIG • -- Word2vec; +PS:I +PS:I • -- GloVe; '&3; '&3; '&3; • --- CBOW;7OMT+VEQ 7OMT+VEQ *EWX8I\X 7OMT+VEQ • -- SkipGram; *EWX8I\X Bunday texnologiyalar hisoblash mashinalarida so‘zlarni ma’no jihatidan barcha *EWX8I\X • -Bunday FastText. texnologiyalar hisoblash mashinalarida so‘zlarni jihatidanifodalash barcha xususiyatlarini ifodalovchi qiymatlarni umumlashtiruvchi vektorma’no ko‘rinishida Bunday texnologiyalar hisoblash mashinalarida so‘zlarni ma’no jihatidanifodalash barcha xususiyatlarini ifodalovchi qiymatlarni umumlashtiruvchi vektor ko‘rinishida Bunday texnologiyalar hisoblash mashinalarida so‘zlarni ma’no jihatidan barcha xususiyatlarini uchun qo‘llaniladi. xususiyatlarini ifodalovchi qiymatlarni umumlashtiruvchi vektor ko‘rinishida ifodalash uchun qo‘llaniladi. ifodalovchi qiymatlarni umumlashtiruvchi vektor ko‘rinishida ifodalash uchun qo‘llaniladi. uchun qo‘llaniladi. Xulosa Xulosa sifatida shuni aytish mumkinki, so‘zning vektor shaklini hosil qilish uchun uning shakl va ma’no jihatidan kuzatilishi mumkin bo‘lgan barcha xusususiyatlari inobatga olishi zarur. Shunda, vektor qiymatlari o‘zaro yaqin bo‘lgan so‘zlar ma’no jihatidan ham bir biriga yaqin so‘zlar bo‘ladi. Hozirda ko‘plab tizimlarda qidirish, guruhlash va bashorat qilish algoritmlarini ishlab chiqishda ushbu yondashuv keng qo‘llanilmoqda. o‘zbek tilidagi so‘zlarning ham shunday vektor korpusini ishlab chiqish kompyuter lingvistikasi va Tabbiy tilni qayta ishlash sohasidagi ko‘plab muammolarga yechim topish imkoniyatini yaratadi. 27

Volume 1, Issue 1 April 2024 28 Authors Elmurod Urinov Department of Communication and Digital Technologies, “University of Management and Future Technologies” universiteti, Tashkent, 100208, Uzbekiston; Nurbek Kholmurodov Department of Communication and Digital Technologies, “University of Management and Future Technologies” universiteti, Tashkent, 100208, Uzbekiston; Correspondence: u.elmurod1988@ gmail.com; Management and Future Technologies journal.umft.uz A COMPARATIVE ANALYSIS OF TRADITIONAL AND DEEP LEARNING-BASED APPROACHES FOR FACE DETECTION: PERFORMANCE, ADVANCEMENTS, AND CHALLENGES A bstract: This thesis presents a comprehensive survey and analysis of various neural network architectures utilized in the domain of face detection, focusing specifically on their effectiveness in handling occlusion. The research investigates the challenges posed by occlusion in face detection and examines state-of-the-art neural network models designed to address this issue. The study explores the strengths, limitations, and advancements made in these architectures concerning occlusion handling. Keyword: Face detection, traditional approaches, deep learning, Viola-Jones, CNN, Copyright: © 2024 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/ by/4.0/).

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Introduction Face detection, a pivotal component of computer vision technology, plays a fundamental role in identifying and localizing human faces within images or video frames. Its ubiquitous relevance spans across diverse domains such as security surveillance, entertainment applications, healthcare diagnostics, personalized marketing, and human-computer interaction. While the significance of accurate face detection cannot be overstated in these realms, the efficacy of this technology encounters numerous hurdles, especially in scenarios where faces are partially obscured or occluded. The evolution of face detection algorithms has significantly enhanced the precision and efficiency of identifying faces in varying environmental conditions and contexts. Despite these advancements, the detection of faces under occlusion remains a persistent challenge, impeding the technology's reliability in real-world applications. Occlusion, stemming from varied factors like objects, hands, facial accessories, or even natural facial expressions, poses a substantial obstacle to accurate face detection. Addressing occlusion in face detection is imperative for the technology's robustness and practical utility across domains. Overcoming occlusion demands the development and deployment of sophisticated algorithms and neural network architectures capable of discerning and reconstructing obscured or partially visible facial features. This survey paper endeavors to comprehensively explore the landscape of face detection under occlusion. It aims to analyze and evaluate a spectrum of neural network architectures and methodologies specifically designed to mitigate the challenges posed by occlusion. By synthesizing existing research and advancements, this paper aims to provide insights into the strengths, limitations, and future directions in the quest for more effective face detection under occlusion. Exploring Face Detection Techniques 2.1 Traditional approaches The Viola-Jones algorithm, pioneered by Paul Viola and Michael Jones, stands out for its adeptness in object detection [1]. It functions through the utilization of Haar-like features, which are rectangular filters designed to detect pixel intensity variations across various sections of an image. These features collectively form a potent classifier capable of efficiently distinguishing between facial and non-facial elements. Employing a cascade of classifiers, the algorithm strategically narrows down regions likely to contain faces, thus notably decreasing computation time. The HOG (histogram oriented gradient) algorithm, developed by Navneet Dalal and Bill Triggs, relies on the distribution of gradient orientations in localized regions of an image. It computes histograms of gradient orientations, which are representations of the local edge directions, and then creates a descriptor capturing the distribution of these gradients. HOG is robust to changes in illumination and background clutter, making it effective for face detection [2]. Each of these traditional algorithms has its strengths. Viola-Jones is known for its speed, HOG for its robustness against variations in lighting and background, and Haar-like features for their simplicity and effectiveness in identifying distinctive patterns in an image. However, they also have limitations. Viola-Jones can sometimes struggle with pose variations and complex backgrounds. HOG, while robust, might not perform as well with extreme variations in facial expressions or occlusions. Haar-like features might oversimplify certain facial characteristics, limiting their accuracy in complex scenarios. These traditional techniques laid the groundwork for more advanced and robust face detection methods by providing insights into feature extraction and classification. Modern approaches often integrate deep learning models, leveraging convolutional neural networks (CNNs) and other sophisticated architectures to achieve higher accuracy and adaptability in various conditions. 29

Volume 1, Issue 1 30 April 2024 Management and Future Technologies journal.umft.uz 2.2 Neural networks based approaches Multi-Scale Fully Convolutional Network (MS-CNN). This is a specialized neural network architecture tailored for precise face detection across multiple scales within an image canvas. This architecture's core strength lies in its ability to handle faces appearing at various sizes and resolutions. By employing a fully convolutional structure, the MS-CNN operates on different scales of the input image, allowing it to detect faces at varying sizes simultaneously. This network architecture is designed to extract hierarchical features from different layers, capturing facial characteristics at multiple levels of abstraction. This comprehensive feature extraction process enables the network to discern faces amidst diverse scales and complexities. Notably, the MS-CNN integrates contextual information from these varied scales, improving its accuracy even in scenarios involving cluttered backgrounds or occlusions that might obscure parts of the face [3]. Typically leveraging an anchor-based approach, the MS-CNN predicts bounding boxes at different scales and aspect ratios across the image, aiding in precisely localizing potential face regions. During training, the network learns to optimize its parameters by minimizing the disparity between predicted bounding boxes and annotated face locations within labeled datasets. MS-CNN has demonstrated promising performance in face detection tasks, exhibiting high accuracy across various benchmarks. Its multi-scale approach addresses the challenge of identifying faces regardless of their sizes or orientations, making it suitable for real-world applications where faces exhibit diverse perspectives and appearances. This architecture represents a significant stride in advancing the accuracy, efficiency, and adaptability of face detection models under varied conditions. Face Detection Algorithm Based on Double-Channel CNN. In literature [4] proposed a model architecture based on VGGNet and residual network. The main reason for choosing VGGNet in this paper is its excellent performance in image feature extraction, and it is easy to modify and train. Residual network is used because it is easy to optimize and can extract face features well with a simple modification. The overall framework of the model is shown in Figure 1. The occlusion perceptron network is used to extract facial features in fewer occlusion regions. The main function of the Region Decision Unit (RD-Unit) is to determine whether the occlusion ratio of the subregion exceeds the set value. When the occlusion of a certain area exceeds the set proportion, the feature vector of the area is discarded. Several regional occlusion judgment units are integrated into the VGG16 network to screen out the subregions with less occlusion. Therefore, the occlusion perceptron network can extract the face related features in fewer occlusion areas. Residual neural network (RNN) is applied to extract full-face features. The output of the two networks is fused with feature vectors by the single-factor weighting method, and the fused feature vectors will be used for face detection. Multi-task Cascaded Convolutional Networks (MTCNN) is a popular deep learning-based algorithm used for face detection. It's a multi-stage framework designed to detect faces in images while handling variations in pose, scale, and lighting conditions. MTCNN consists of three stages, each focusing on a specific aspect of face detection [5]: Stage 1: Proposal Network (P-Net). In this stage generates candidate facial regions likely to contain faces. It employs a fully convolutional network to scan the image at multiple scales, detecting faces of different sizes. Utilizes a sliding window technique to propose candidate bounding boxes with associated probabilities for face regions. Works efficiently in identifying potential face locations but might produce false positives; Stage 2: Refinement Network (R-Net). In this stage refines the candidate face regions obtained from the first stage. Filters out false positives by performing more precise bounding box regression. Evaluates the proposed regions, discarding inaccurate regions and further narrowing down potential face regions. Improves the accuracy of face localization by refining the bounding boxes;

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Stage 3: Output Network (O-Net). The final stage performs further refinement and detailed facial feature extraction. Detects facial landmarks (such as eyes, nose, and mouth) within the bounding boxes identified by the previous stages. Outputs the final bounding boxes for faces along with the precise facial landmarks. MTCNN is widely used in applications requiring accurate and robust face detection, such as facial recognition systems, surveillance, and biometric authentication. However, the choice of a face detection algorithm depends on factors like accuracy, speed, and resource constraints specific to the intended application. Conclusion In the domain of face detection, traditional methods like Viola-Jones and Histogram of Oriented Gradients (HOG) offer simplicity and speed but struggle with handling diverse conditions. Contrastingly, neural network-based algorithms such as Multi-Stage Cascaded Convolutional Networks (MTCNN), Double-Channel CNN, and Multi-Scale Convolutional Neural Network (MS-CNN) provide robustness and accuracy, leveraging deep learning techniques to address variations in scale, pose, and lighting. However, their effectiveness often comes at the cost of computational intensity and the need for substantial labeled datasets for training. Each approach presents a trade-off between speed, accuracy, and resource requirements, catering to specific application needs within face detection tasks. References 1. Viola, P.; Jones, M. (2001). “Rapid object detection using a boosted cascade of simple features”. Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001. Vol. 1. IEEE Comput. Soc. 2. William T. Freeman, Michal Roth, “Orientation Histograms for Hand Gesture Recognition”, Tech. Rep. TR94-03, Mitsubishi Electric Research Laboratories, Cambridge, MA, December 1994. 3. S. R. Dubey, “Local directional relation pattern for unconstrained and robust face retrieval,” Multimedia Tools and Applications, vol. 78, no. 19, pp. 28063–28088, 2019. 4. H. Li, Z. Lin, X. Shen, J. Brandt and G. Hua, “A convolutional neural network cascade for face detection,” 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 2015, pp. 5325-5334, doi: 10.1109/CVPR.2015.7299170. 5. J. Xiang and G. Zhu, “Joint Face Detection and Facial Expression Recognition with MTCNN,” 2017 4th International Conference on Information Science and Control Engineering (ICISCE), Changsha, China, 2017, pp. 424-427, doi: 10.1109/ICISCE.2017.95. 31

Volume 1, Issue 1 April 2024 32 Mualliflar Farrux Axmedov Kommunikatsiya va raqamli texnologiyalar kafedrasi, “University of Management and Future Technologies” universiteti, Toshkent, 100208, O‘zbekiston; Muhriddin Umarov Kommunikatsiya va raqamli texnologiyalar kafedrasi, “University of Management and Future Technologies” universiteti, Toshkent, 100208, O‘zbekiston; Mas’ul: axmedovfarrux@336gmail. com Management and Future Technologies journal.umft.uz O‘ZBEKCHA MATNDAN NUTQ SIZTEZLASH DASTURIY TIZIMI UCHUN DIFONLAR BAZASININING QURILISHI A nnotatsiya: Mazkur ishda o‘zbekcha matndan nutq sintez qilish dasturi uchun kiritilgan matnni tahlil qilish, fonetik qoidalar asosida difonli bo‘lakarga ajratish va ushbu bo‘laklar uchun ma’lumotlar bazasini shakllantirish tamoyillari ko‘rib chiqiladi. Kalit so‘zlar: matn, nutq, sintezlash, difon, baza, unli, undosh. Copyright: © 2024 by the authors. Ushbu maqola Creative Commons Attribution (CC BY) litsenziyasi shartlari asosida tarqatiladigan ochiq foydalanish maqolasi hisoblanadi (https://creativecommons.org/ licenses/by/4.0/).

Management and Future Technologies journal.umft.uz CONSTRUCTION OF A DIPHONE BASE FOR A UZBEKI TEXT-TO-SPEECH SOFTWARE SYSTEM A April 2024 Volume 1, Issue 1 33 Authors Farrukh Akhmedov Department of Communication and Digital Technologies, University of Management and Future Technologies, Tashkent, 100208, Uzbekistan; Muhriddin Umarov bstract: In this work, the principles of analyzing the text entered for the Uzbek text-to-speech synthesis program, dividing it into diphonic fragments based on phonetic rules, and forming a database for these fragments are considered. Department of Communication and Digital Technologies, University of Management and Future Technologies, Tashkent, 100208, Uzbekistan; Keywords: text, speech, synthesis, diphone, base, vowel, consonant. Correspondence: akhmedovfarrukh@336gmail.com Copyright: © 2024 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/ by/4.0/).

Volume 1, Issue 1 34 April 2024 Management and Future Technologies journal.umft.uz Kirish Matndan nutq sintezlash (MNS) tizimi – bu berilgan matnni nutq tovushlariga o‘tkazish tizimidir. Bunda matn manbaasi turlicha bo‘lishi mumkin. Shu o‘rinda MNS tizimi matnni qayta ishlash va nutq yaratish qismlarini o‘z ichiga qamrab olgan tizimdir. Ideal MNS tizim deganda inson nutqini aniq va to‘liq talaffuz eta oluvchi tizim tushuniladi. Misol uchun u raqamlarni va qo‘lyozma matnlarni o‘qiy olishi lozim, so‘zlarni to‘g‘ri talaffuz eta olishi zarur va hokazo. Matnni tahlil qilish qismi ideal tizimning eng murakkab qismidir, chunki, unda muayyan matnni tushunishga va aniq natija olishga buyurtma beriladi, matn esa o‘ta aniqlikda grammatik tahlil qilinadi. Matnni tahlil qilish qismining asosiy vazifasi – kiritiluvchi matnni nutq hosil qilish qismiga yuborish uchun “kichik qismlar”ga ajratishdir. Matnni tahlil qilish qismining ikkinchi vazifasi – so‘z va gapdagi talaffuzni aniqlashdir. Bunda ma’lumot nutq hosil qilish qismiga u tushunadigan formatda yuboriladi. Matn tahlili qismining bunday vazifalari ko‘p jihatdan MNS tizimining sifat darajasini belgilab beradi. Matnni qayta ishlash. Ushbu tizimning matnni qayta ishlovchi qismi dastlab, berilgan matnni so‘zlarga, so‘ngra so‘zlarni difon deb ataluvchi bo‘laklarga ajratadi. Difon – so‘zning ikki yoki uch harfidan iboratbo‘linishi nutqiy bo‘lagi bo‘lib, u tizimning sintez qilish tomonidan qabul qilinadi bo‘laklarga kerak. Agar bundaynutqni bo‘laklashni uchqismi hafrdan iborat chegarani va qayta ishlanadi. Tizimning nutqni qayta qismi to‘g‘ri va aniq ishlashi uchun so‘zlar mumkin saqlagan holda bajarib bo‘lmasa (ya’ni,ishlash ketmaOIXYRHSWLLEVJPEVOIPWE YLSPHENEVERKWM^ qadar jarangli (unli harf qatnashgan) bo‘laklarga bo‘linishi kerak. Agar bunday bo‘laklashni uch hafrdan qismlarda ham bo‘laklash amalga oshiriladi. iborat chegarani saqlagan holda bajarib bo‘lmasa (ya’ni, ketma-ket undosh harflar kelsa), u holda jarangsiz qismlarda ham bo‘laklash amalga oshiriladi. %WSWM]UMWQ 2YXU LSWMP UMPMWL UMWQMRMRK ZE^MJEWM – XSZYWLRM MQOSR UEHEV QSWPEWLXMVMF XSZYWL GLMUEVMWL UYVMPQEWMKE Y^EXMWLHER MFSVEX %KEVqism LEVJPEV WSRM PMQMXHER SWLMF OIXWE Y LSPHE Asosiy NEVERKWM^ bo‘lakchalarda ham bo‘laklanish amalgaqadar oshiriladi. Jarangli Nutq hosil qilish qismining vazifasi – tovushni imkon moslashtirib, tovush qismlarni chiqarish FMVPEWLXMVMWL YGLYR 4730% EPKSVMXQMHER .EVERKWM^ FIZSWMXE qurilmasiga uzatishdan iborat. Agar harflar soniJS]HEPERMPEHM limitdan oshib ketsa, u holdaUMWQPEV jarangsizIWE bo‘lakchalarda to‘g‘ridanto‘g‘ri AgarJarangli unli harf ikkita undosh harflar o‘rtasida kelsa, u holda ham bo‘laklanish birlashtiriladi. amalga oshiriladi. qismlarni birlashtirish uchun PSOLA algoritmidan bu unli harf har ikkala difonga qo‘shiladi. Masalan, “fan” so‘zi “fa” va “an” kabi bo‘linadi. Bu foydalaniladi. Jarangsiz qismlar esa bevosita, to‘g‘ridan-to‘g‘ri birlashtiriladi. Agar unli harf ikkita yerda “a” harfio‘rtasida har ikkala kiritilgan. bunday bo‘laklashni qismlar undosh harflar kelsa,difonga u holda bu unli harfLekin har ikkala difonga qo‘shiladi. jarangsiz Masalan, “fan” so‘zi uchun ishlatib bo‘lmaydi. Ideal holatda barcha bo‘laklar jarangli bo‘lishi kerak. Biroq, “fa” va “an” kabi bo‘linadi. Bu yerda “a” harfi har ikkala difonga kiritilgan. Lekin bunday bo‘laklashni bo‘laklardagi sonining 45 tadan oshib barcha ketishibo‘laklar muammo tug‘diradi. Masalan, jarangsiz qismlarharflar uchun ishlatib bo‘lmaydi. Ideal holatda jarangli bo‘lishi kerak. Biroq, “shartli” so‘zi quyidagicha bo‘laklarga bo‘linishi kerak: “sha” va “artli”. Ushbu birikmaning 2 bo‘laklardagi harflar sonining 4-5 tadan oshib ketishi muammo tug‘diradi. Masalan, “shartli” so‘zi bo‘lagidagi harflar soni 5 ta. Bunday vaziyatda ikkinchi bo‘lak harflari ikki bo‘lakka quyidagicha bo‘laklarga bo‘linishi kerak: “sha” va “artli”. Ushbu birikmaning 2-bo‘lagidagi harflar soni 5 ajratiladi. Jarangsiz qismlar odatda, turg‘un qismi bo‘lib,qismlar ularniodatda, birlashtirishda ta. Bunday vaziyatda ikkinchi bo‘lak harflarinutqning ikki bo‘lakka ajratiladi. Jarangsiz nutqning sifati deyarli pasaymaydi. Tizimda so‘zlarning bo‘laklarga bo‘lib olinishini ushbu jadval turg‘un qismi bo‘lib, ularni birlashtirishda sifati deyarli pasaymaydi. Tizimda so‘zlarning bo‘laklarga SVUEPM]E\WLMVSUXYWLYRMFSPMWLQYQOMR bo‘lib olinishini ushbu jadval orqali yaxshiroq tushunib olish mumkin. NEHZEPTizimda so‘zlarning bo‘laklarga bo‘lib olinishini. 1-jadval. Tizimda so‘zlarning bo‘laklarga bo‘lib olinishini. /-6-8-7, ',-5%6-7, (EWXYV (EEWXYYV 7EQEVE 7EEQEEVE 8Y^MPQE 8YY^MMPQE 7EVS] 7EEVSS] 7SEX 7SEX &MVPEWLQE FMMVPEEWLQE Kiritiluvchi matn faqat harflar va sonlardan tashkil topgan bo‘lishi lozim. Sonlar tizimda so‘zlarga Kiritiluvchi harflar va sonlardan lozim. Sonlar aylantiriladi. Tizimmatn tinishfaqat belgilarini tushirib qoldiraditashkil hamdatopgan tizimdabo‘lishi aniq vaqt belgilanmaydi. tizimda so‘zlarga aylantiriladi. Tizim tinish belgilarini tushirib qoldiradi hamda tizimda aniq ZEUXFIPKMPERQE]HM Ma’lumotlar bazasining shakllantirilishi. Ma’lumotlar bazasini shakllantirish – 127 tizimining eng zaruriy qismlaridan biri bo‘lib, undan nutq sintez qilish qismida mos tovush fayllarini olish uchun foydalaniladi. Birlashyitishlarda qo‘llaniluvchi difonli bo‘lakchalarning

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Ma’lumotlar bazasining shakllantirilishi. Ma’lumotlar bazasini shakllantirish – MNS tizimining eng zaruriy qismlaridan biri bo‘lib, undan nutq sintez qilish qismida mos tovush fayllarini olish uchun foydalaniladi. Birlashyitishlarda qo‘llaniluvchi difonli bo‘lakchalarning tovush fayllari tizim ishga tushmasidan avval, ma’lumotlar bazasiga joylashtirilgan bo‘lishi lozim. Ulardan nutq sintezi qismida nutqni moslashtirish jarayonida foydalaniladi. Kiritilgan matndan ajratib olingan bo‘laklar uchun mos difonlarni ma’lumotlar bazasidan olish mumkin. Matndan yoki to‘liq so‘zdan difon olish, yaxshi talaffuzli nutq hosil qilishda tayyor difonlardan foydalanishdan ko‘ra yaxshiroq sifat ko‘rsatkichini beradi. Matn yoki so‘z kiritishni hal qilish muhim bosqich. Bir qancha mulohazalardan so‘ng shu narsa ma’lum bo‘ldiki, yangi yozilgan ibora ma’lumotlar bazasiga ko‘pgina difonlarni qo‘shadi. Bunda, difonlarning aksariyat qismi oldindan yozib qo‘yilgan bo‘ladi. Shunga qaramasdan, yozilgan difonlar soni unga mos difonlar sonining 10% ini tashkil etadi. Shuning uchun biz difon olishning yaxshiroq yo‘lini topishimiz lozim. Maqsad – bazaga yangi difon kiritmasdan, so‘zlar ustida ishlashga ko‘p vaqt sarflamaslik uchun mumkin qadar kamroq so‘zlarni yozishdir. Kiritilgan so‘zlarda mumkin qadar takrorlanuvchi difonlar soniga mos holda yozilgan so‘zlar soni qisqartiriladi. Kiritilgan matnda buni bajarish juda qiyin bo‘lganligi sababli dasturda shu ishni bajaruvchi qism dastur kritiladi. Dastur matndagi barcha so‘zlar ro‘yxatini tuzadi va ularga mos bo‘lgan difonlar sonini hisoblaydi, so‘ngra, bu so‘zlardan olingan difonlarni ro‘yxatga oladi va ro‘yxatda yo‘q bo‘lgan difonlar sonini barcha so‘zlardan tekshirib chiqadi. Eng ko‘p difonlarni qabul qilgan so‘z olinadi va yana jarayon shu tarzda davom etadi. Jarayon barcha so‘zlarda yangi difonlar yo‘qolgunga qadar davom etadi. Biroq, so‘zlar soni oshib ketganda operatsiyani bajarish uchun juda ko‘p vaqt talab etiladi. Ushbu muammoni hal etish uchun boshqacharoq yo‘sinda yondoshish kerak. Yaratilgan so‘zlar ro‘yxatidagi harflar soni 16 tadan ko‘p bo‘lgan so‘zlarning ixchamroq ro‘yxati yasaladi. Jarayon bu ro‘yxatda qisqa vaqt ichida bajariladi. Agar barcha difonlarni olish imkoni bo‘lmasa, limit qisqartiriladi va jarayon yana davom ettiriladi. Ushbu metodlardan foydalanib, kiritilgan matn yoziladi. Tovush ohangi so‘zning turli joyida turlicha eshitilishi mumkin. So‘z tarkibidan olingan tovush – shunga o‘xshash boshqa qismdan olingan tovush bilan birgalikda kelishi nuqtayi nazaridan, ba’zi o‘zgarishlar bo‘lish ehtimolligi oshadi, chunki, fonemalar bo‘yicha harakatlanadi. Shuning uchun, yaxshiroq nutq yasash maqsadida, so‘zning turli qismlari uchun turli difonlar faylga yozilgan bo‘lishi kerak. Bu qismlarni uchga bo‘lish mumkin: boshlang‘ich qism, o‘rta qism va so‘ngi qismlar. Ushbu qismlar uchun turli difonlarni yozib olinadi va ulardan mos ravishda foydalaniladi. Bunda so‘z yasashda so‘zning birinchi qismi uchun ajratib olingan difon o‘rta qismi yoki so‘ngi qismida ishlatilmaydi. Masalan, agar ”da” difoni “daftar” so‘zidan olib yozilgan bo‘lsa, u “oldida” sintezida qo‘llanilmaydi. Mazkur tamoyil asosida, difonlar va harflarni farqlash tamoyili ishlab chiqildi. Unga ko‘ra agar “a” harfi so‘z oxirida kelsa, tizim difonlarni o‘qishda “a” dan keyin “_” belgisini qo‘yib ajratadi. Masalan, “no” difoni “nomli” so‘zidan ajratib olingan bo‘lsa, u “nomli” so‘zining boshlang‘ich qismida kelganligi uchun “no_b” (boshlang‘ich) kabi yoziladi, agar “no” difoni “asnoda” so‘zidan ajratib olingan bo‘lsa, “no” so‘zning o‘rtasida kelganligi uchun “no_o” kabi yoziladi va agar, “no” difoni “istisno” so‘zidan ajratib olingan bo‘lsa, “no” so‘zning so‘ngida kelganligi bois “no_s” kabi yoziladi. Har bir difon o‘zining o‘ng va chap taraflarining har ikkalasi uchun tebranish qiymatlariga ega. Agar chap yoki o‘ng tomon jarangsiz bo‘lsa, u holda, bu qism bo‘yicha tebranish qiymati 0 (nol)ga keltiriladi. Masalan, hosil qilingan ma’lumotlar bazasida “ka_b” bo‘yicha tebranish qiymati chap tomon uchun 0 ga va o‘ng tomon uchun 147 ga teng; “ara_o” uchun tebranish qiymati chap tomon uchun 142 ga va o‘ng tomon uchun 145 ga teng. Real vaqtda bu qiymatlardan biri hisoblanishi kerak yoki birgalikda hisoblanib, ishga tushiriladi. Tizim ishlayotgan jarayonda buni avtomatik tarzda bajarish qiyin bo‘lganligi sababli, uni ma’lumotlar bazasini shakllantirish vaqtida hisoblagan ma’qul. Yuqoridagi fikr mulohazalardan kelib chiqib, difonli ma’lutlar bazasini shakllantirishda quyidagicha maydonlarning mavjudligi hisobga olinishi lozim: 35

o‘ng tomon uchun 145 ga teng. Real vaqtda bu qiymatlardan biri hisoblanishi kerak yoki FMVKEPMOHELMWSFPERMFMWLKEXYWLMVMPEHM8M^MQMWLPE]SXKERNEVE]SRHEFYRMEZXSQEXMOXEV^HE bajarish qiyin bo‘lganligi sababli, uni ma’lumotlar bazasini shakllantirish vaqtida hisoblagan ma’qul. Yuqoridagi fikr mulohazalardan kelib chiqib, difonli ma’lutlar bazasini Volume 1, Issue 1 April 2024 Management and Future Technologies journal.umft.uz WLEOPPERXMVMWLHEUY]MHEKMGLEQE]HSRPEVRMRKQEZNYHPMKMLMWSFKESPMRMWLMPS^MQ 36 NEHZEPDifonli ma’luotlar bazasining ma’lumot maydonlari 2-jadval. Difonli ma’luotlar bazasining ma’lumot maydonlari &SWLPER O‘rta So‘ng 8IFVERMWL 8IFVERMWL 8 Boshlang‘ich So‘nggi diapazon g‘ich HMJS M UM]QEXM qiymati o‘ng V HMETE^SR HMJSR R HMJSR GLETXSQSR XSQSR EECF EECS EECW EFCF EFCS EFCW IOMC IOMCF IOMCW S NEHZEPDifonli ma’luotlar bazasining ma’lumot maydonlari Ushbu jadval misol sifatida keltirilgan. Undagi “Boshlang‘ich diapazon” hamda “So‘nggi diapazon” deb nomlangan maydonlarda mazkur difonli bo‘lakchaning oldindan yozib olingan nutq fayli tarkibidagi boshlang‘ich va so‘nggi nuqtalardagi diapazon qiymati. Bu ustundagi qiymatlar difonning boshlang‘ich, o‘rta va so‘nggi qismlari uchun slesh (“/”) belgisi bilan ajratib ko‘rsatilgan. Ma’lumotlar bazasini shakllantirishda, dasturlash jarayoni hamda dasturning bajarilish jarayonlaridagi ortiqcha hisoblashlarning oldini olish maqsadida, ularni alohida maydonlarda kiritish kerak bo‘ladi. Xulosa Matndan nutq sintezlashning keng ommalashgan usullaridan biri – konkatenativ sintezlash uslubidir. Bunda difonli bo‘laklarning bog‘lanishi asosida yangi nutqning sintezlanishi tamoyili qo‘llaniladi. Ishda o‘zbekcha matndan nutq sintez qilish dasturi uchun kiritilgan matnni tahlil qilish, fonetik qoidalar asosida difonli bo‘lakarga ajratish va ushbu bo‘laklar uchun ma’lumotlar bazasini shakllantirish masalalari ko‘rib chiqildi. Bunda difonlar bazasini shakllantirish va bu bosqichda o‘zbekcha matnlar uchun yuzaga keluvchi ba’zi muammolar va ularning yechimlari xususida fikrlar yoritildi. Adabiyotlar 1. Allen J., Hunnicutt S., Klatt D. From Text to Speech: The MITalk System. Cambridge University Press, Inc. 2. Taylor P. – “Text-To-Speech-Synthesis”, University of Cambridge, 2007. 3. Б. М. Лобанов, Л. И. Цирульник «Компьютерный синтез и клонирование речи». — Минск, «Белорусская Наука», 2008. — 316 стр. 4. B.Bahriddinova, “Hozirgi o`zbek tili fonetika-fonologiyasi” – ma’ruzalar matni, Qarshi Davlat Universiteti, 2007-y.

Management and Future Technologies journal.umft.uz NUTQNI TANIB OLISHNI AMALGA OSHIRISH BOSQICHLARI A nnotatsiya: Ushbu maqolada nutq signalini tanib olishning sohalarga tadbiqlari, buyruqlarni tanib olish bosqichlari, nutq signalini tahlil qilish algoritmlari, ularda qo‘llaniladigan uslublar va yondashuvlar yoritib berilgan. April 2024 Volume 1, Issue 1 37 Muallif O‘rinboy Naziraliyev Kommunikatsiya va raqamli texnologiyalar kafedrasi, “University of Management and Future Technologies” universiteti, Toshkent, 100208, O‘zbekiston; naziraliyevurinboy1996@gmail.com Kalit so‘zlar: nutqni tanish, Smart House, IoT (Internet buyumlari), formant chastota, fragment, spektral tahlil, xususiyatlarni ajratib olish, kNN, SVM, Markov zanjiri, akustik ishlov berish, ovoz modeli, filtrlash. Copyright: © 2024 by the authors. Ushbu maqola Creative Commons Attribution (CC BY) litsenziyasi shartlari asosida tarqatiladigan ochiq foydalanish maqolasi hisoblanadi (https://creativecommons.org/ licenses/by/4.0/).

Volume 1, Issue 1 April 2024 38 Author Urinboy Naziraliev Department of Communication and Digital Technologies, University of Management and Future Technologies, Tashkent, 100208, Uzbekistan; naziraliyevurinboy1996@gmail.com Management and Future Technologies journal.umft.uz THE MAIN PROCESS TO IMPLEMENT SPEECH RECOGNITION A bstract: In this article, field applications of speech signal recognition, command recognition steps, speech signal analysis algorithms, methods and approaches used in them are highlighted. Keywords: speech recognition, Smart House, IoT (Internet of Things), formant frequency, fragment, spectral analysis, feature extraction, kNN, SVM, Markov chain, acoustic processing, sound model, filtering. Copyright: © 2024 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/ by/4.0/).

Management and Future Technologies journal.umft.uz April 2024 Volume 1, Issue 1 Kirish Nutqni tanish – bu nutq signalini tanib olinishi kerak bo‘lgan so‘zlar yoki harflarga aylantirishning statistik jarayonidir. Texnologiya yutuqlarining rivojlanishi bilan nutq so‘zlarini tanish quyidagi sohalarda keng qo‘llanilmoqda: 1. Xavfli xudud va yuqori zararga ega bo‘lgan joylarda sanoat dastgohlarini nutq buyruqlari orqali masofadan boshqarish. 2. “Smart House” tizimlarida uy jihozlarini boshqarish, foydalanuvchini aniqlash va nazorat qilish. 3. Transport vositalarini boshqarishda – shtat va bort tizimlarini ovozli boshqarish (multimedia tizimlari, navigatsiya qurilmalari, bort kompyuterlari va h.k.). 4. IoT (ing. Internet of things) texnologiyalarini amalga oshirishda – masofadan boshqarish va identifikatsiya qilishda. 5. Imkoniyati cheklangan insonlar uchun interfeys – buyruqlarni bajarish, ovozli boshqarish, ovozli assistent, nutqni matnga o‘girish. 6. Xizmat ko‘rsatish sohalarida – infokioskalarda ma’lumot olish, “Call Center” larda avtomatik javob berish, mijozlar xohishlarini o‘rganish tizimlari. Kundan kunga real vaqtda nutqni tanib olish va tahlil qilish tizimlarga talab ortib bormoqda. Bunday real vaqt tizimlari orqali katta ma’lumotlar bazasi talablari, nutq signallarining shovqinga ta’sirchanligi va yuqori hisoblash resurslari talablari bo‘lganligi uchun o‘rnatilgan apparat-dasturiy platformada boshqarish qiyin. Qayta dasturlanuvchi integral sxemalar(ing. Field Programmable Gate Array) va maxsus mobil protsessorlarga asoslangan apparat-dasturiy platformalar bunday real vaqt tizimlarining samarasani oshiradi va nutq buyruqlarini tanib olish imkoniyati beradi. Bunday platformalar oson dasturlash imkoniyati va yuqori hisoblash samaradorligiga ega ekanligi bilan nutq signallariga ishlov berishda munosib platforma hisoblanishi mumkin. Nutq signali – bu murakkab dinamik jarayonning modeli bo‘lib, uni tahlil qilishda signal va uning fragment qismini tavsiflovchi bir nechta ko‘rsatikich (xususiyatlar)lariga tayaniladi. Nutq signallarining ushbu asosiy xususiyalari quyidagilar: formant chastotasi, asosiy ton chastotasi, spektral tashkil etuvchilardir. Asosiy qism Tahlil qilish uchun nuqt signalining eng sodda ko‘rinishi, bu ajratib olingan so‘zlardir. Odatda so‘zlar va ularning lug‘atlari cheklangan bo‘ladi. Zamonaviy murakkab nutqni tanish tizimlarida ham asosan o‘zak so‘zlarni ajratib olish orqali amalga oshiriladi. Nutq buyruqlari orqali boshqarish tizimlarida esa juda ham qo‘l keladi. Roman-german oilasiga mansub va ba’zi bir osiyo tillar uchun nutqni tanib olish algoritmlari va modellari ishlab chiqilgan. Lekin, ushbu tillar uchun ishlab chiqilgan model va algoritmlar “o‘zbek tili” uchun to‘g‘ri kelmaydi. Bundan tashqari ularni apparat-dasturiy vositalarida amalga oshirish hozirgi kunda ham aktual hisoblanib, maxsus yondashuvlarni talab etadi. Nutqni avtomatik tanib olish tizimlari quyidagi bosqichlarda amalga oshiriladi (1-rasm): xususiyatlarni ajratib olish, o‘qitish va tanib olish. 39

EPKSVMXQPEVM ZE QSHIPPEVM MWLPEF GLMUMPKER 0IOMR YWLFY XMPPEV YGLYR MWLPEF GLMUMPKER model va algoritmlar “O‘zbek tili” uchun to‘g‘ri kelmaydi. Bundan tashqari ularni apparat HEWXYVM] ZSWMXEPEVMHE EQEPKE SWLMVMWL LS^MVKM OYRHE LEQ EOXYEP LMWSFPERMF QE\WYW ]SRHEWLYZPEVRMXEPEFIXEHM Volume 1, Issue 1 April 2024 Management and Future Technologies journal.umft.uz 2YXURM EZXSQEXMO XERMF SPMWL XM^MQPEVM UY]MHEKM FSWUMGLPEVHE EQEPKE SWLMVMPEHM rasm): xususiyatlarni ajratib olish, o‘qitish va tanib olish. 40 1-rasm. Nutq tanibVEWQ2YXUXERMFSPMWLRMEQEPKESWLMVMWLFSWUMGLPEVM olishni amalga oshirish bosqichlari (EWXPEFOM MWLPSZ FIVMWL – WMKREPRM EREPSKHER ko‘rinishga Dastlabki ishlov berish – signalni qabul UEFYP qilish, UMPMWL analogdan raqamli raqamli ko‘rinishga o‘zgartirish, o‘zgartirish, filtrlashdan iborat. filtrlashdan iborat. <YWYWM]EXPEVRM ENEVMXMWL – jimlik sohalarini olib tashlash, oynalardan o‘tqazish, Xususiyatlarni ajaritish – jimlik sohalarini olib tashlash, oynalardan o‘tqazish,1IPGLEWXSXEZM] spektr qiymatlarini va WTIOXVUM]QEXPEVMRMZE1*'' MRK1IP*VIUYIRG]'ITWXVEP'SIJJMGMIRX MFCC(ing. Mel Frequency Cepstral Coefficient - Mel chastotaviy kepstral koeffitsienti) parametrlarini OITWXVEPOSIJJMXWMIRXM TEVEQIXVPEVMRMLMWSFPEWLRMXEWLOMPIXEHM hisoblashni tashkil etadi. olish – bunda har bir nutq so‘zi olingan parametralari bo‘yicha O‘qitish va tanib mashinali o‘qitish va sun’iy intellekt algoritmlari orqali o‘qitiladi va tanib olinadi. o‘qitish va tanib olish – bunda har bir nutq so‘zi olingan parametralari bo‘yicha mashinali o‘qitish Nutqni avtomatik tanib olishning muvaffaqiyatlilari ajratilgan so‘zlarni tanib va sun’iy intellekt algoritmlari orqali o‘qitiladieng va tanib olinadi. olish uchun suxandonga sozlangan tizimlar bo‘lib chiqdi. Bunday tizimlar ikki rejimdan Nutqni avtomatik tanib olishning eng muvaffaqiyatlilari ajratilgan so‘zlarni tanib olish uchun birida ishlaydi. O‘qitish rejimida foydalanuvchi tizimga mikrofonda tanib olish uchun har suxandonga sozlangan tizimlar bo‘lib chiqdi. Bunday tizimlar ikki rejimdan birida ishlaydi. o‘qitish bir so‘zni aytib, ovozini tanib olishga o‘rgatadi. Tizim har bir so‘zni rejimida foydalanuvchi tizimga mikrofonda tanib olish uchun har bir so‘zni aytib, ovozini tanib olishga na’munasini(etalonini) yaratadi va xotirada saqlaydi. o‘rgatadi. Tizim har bir so‘zni na’munasini(etalonini) yaratadi va xotirada saqlaydi. Tanib olish rejimida har bir aytilgan so‘z yana raqamlashtiriladi va uning na’muna Tanibxotiradan olish rejimida har birna’muna aytilgan so‘z yana raqamlashtiriladi va uningAgar na’muna etaloni xotiradan etaloni olingan etalonlar bilan taqqoslanadi. mos keladigan olingan unda na’muna etalonlar bilan taqqoslanadi. mosfoydalanuvchini keladigan bo‘lsa, unda so‘z tanib olinadi bo‘lsa, so‘z tanib olinadi va tizim buAgar haqda xabardor qiladi yoki va tizim bu foydalanuvchini xabardor qiladi yoki biron bir MWLPEWLMWY\ERHSRRMRK harakatni amalga oshiradi. Bunday FMVSR FMVhaqda LEVEOEXRM EQEPKE SWLMVEHM&YRHE] XM^MQPEVRMRK LEV FMV tizimlarning suxandonning har bir so‘zshovqinli oldidan yetarlicha qilishiga,aniq shovqinli fonga va so‘z oldidanishlashi yetarlicha pauza qilishiga, fonga vapauza so‘zlarning va to‘g‘ri so‘zlarning aniq va to‘g‘ri talaffuz qilinishiga bog‘liq. Tanib olish paytida nutq signallariga raqamli ishlov talaffuz qilinishiga bog‘liq. Tanib olish paytida nutq signallariga raqamli ishlov berishning eng muhim amallari, xususiyatlarni ajratish, og‘zaki nutqlardan namunalar olinib berishning eng muhim amallari, xususiyatlarni ajratish, og‘zaki alohida nutqlardan alohida etalonlar yaratish vaIXEPSRPEV namunalardan kerakligini tanlab olish, etalonlar xotirada saqlanadigan yozuvlar REQYREPEV SPMRMF ]EVEXMWL ZE REQYREPEVHER OIVEOPMKMRM XERPEF SPMWL IXEPSRPEV bilan taqqoslash bajariladi (2-rasm). \SXMVEHEWEUPEREHMKER]S^YZPEVFMPERXEUUSWPEWLFENEVMPEHM VEWQ VEWQ2YUXRMXERMFSPMWLXM^MQMRMRKYQYQM]FPSOW\IQEWM 2-rasm. Nuqtni tanib olish tizimining umumiy blok-sxemasi 2YXURM XELPMP UMPMWL ZE \YWYWM]EXPEVRM ENVEXMWL FSWUMGLPEVM UY]MHEKMPEVHER XEWLOMP XSTMWLMQYQOMR Nutqni tahlil qilish va xususiyatlarni ajratish bosqichlari quyidagilardan tashkil topishi mumkin: • EOYWXMO •akustik; • JSRIXMO •fonetik; • PIOWMO • WMRXEOXMO •leksik; 2YXU WMKREPPEVMKE MWLPSZ FIVMWLRMRK HEWXPEFOM ZE RMWFEXER IRK WSHHE HEVENEWM FY •sintaktik; akustik va fonetik tahlil bo‘lib, asosan gap bo‘lagini ichidagi alohida ajratilgan so‘zlarni Nutq signallariga ishlov berishning dastlabkiso‘zlarini va nisbatan eng sodda darajasi bu akustik fonetik tanib olishga bag‘ishlangan. Nutq buyruq tanib olishda akustik ishlov va berish tahlil bo‘lib, asosan gap bo‘lagini ichidagi alohida ajratilgan so‘zlarni tanib olishga bag‘ishlangan. Nutq FSWUMGLMRMEQEPKESWLMVMWL]IXEVPMLMWSFPEREHM/I]MRKMRMWFEXERQYVEOOEFHEVENEWMIWE buyruq so‘zlarini tanib olishda akustik ishlov berish bosqichiniqo‘shma amalga oshirish yetarli hisoblanadi. bu leksik, sintaktik va semantik tahlil hisoblanib, murakkab gaplarni Keyingi nisbatan murakkab darajasi esa bu leksik, sintaktik va semantik tahlil hisoblanib, qo‘shma XYWLYRMWLKEUEVEXMPKER murakkab gaplarni tushunishga qaratilgan. &ENEVMPKERMPQM]MWLPERQEPEVZEIOWTIVMQIRXEPXEHUMUSXPEVRMXELPMPUMPMFUEVE]HMKER bo‘lsak, nutqni tahlil qilish va tanib olish usullari va algoritmlari orasida alohida ajartilgan nutqlarga ishlov berishning quyidagi protseduralarini ajratib ko‘rsatishimiz QYQOMR VEWQ %OYWXMO XELPMP QYLMXHE XEVUEPKER EOYWXMO E\FSVSXRM ERMUPEWL XELPMP UMPMWL ZE XEWRMJPEWLKEUEVEXMPKERYWYPPEVRMERKPEXEHM%OYWXMOXELPMPUMPMWLEZXSQEXMOOY^EXYZEUPPM