Zobrazeno 1 - 10
of 111
pro vyhledávání: '"Levent M. Arslan"'
Autor:
I. Rasim Ulgen, Levent M. Arslan
Publikováno v:
2022 IEEE Spoken Language Technology Workshop (SLT).
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::3a5c7ce1718d27c32f4034e0dac6fe14
https://doi.org/10.1109/slt54892.2023.10022331
https://doi.org/10.1109/slt54892.2023.10022331
Autor:
Ayşe Gürel, Levent M. Arslan
Publikováno v:
Dilbilim Araştırmaları Dergisi, Vol 19, Iss 1, Pp 77-90 (2008)
Speech/voice recognition by machines has been a topic of interest since 1950s. Research that initially adopted dynamic programming methodologies now mostly uses the hidden Markov model as the method for speech recognition. Nevertheless, even the most
Externí odkaz:
https://doaj.org/article/ea2d908404c64f7c8bb37d783ffe29e4
Publikováno v:
2021 Innovations in Intelligent Systems and Applications Conference (ASYU).
In this study, customer churn analysis is performed using various deep learning approaches. Customer churn analysis systems try to predict whether customers will continue to use the products or not, and this analysis allows companies to increase thei
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::61740ba63593c83defe5d968dd214945
https://doi.org/10.1109/asyu52992.2021.9598974
https://doi.org/10.1109/asyu52992.2021.9598974
Publikováno v:
2021 Innovations in Intelligent Systems and Applications Conference (ASYU).
In this study, X-Vector and Ecapa-Tdnn models, which are recently used in speaker recognition, are proposed to be used in spoken language detection problem, and their accuracies are compared. Real agent-customer conversations from different call cent
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::844ae294059117c24865b997d368610b
https://doi.org/10.1109/asyu52992.2021.9599040
https://doi.org/10.1109/asyu52992.2021.9599040
Publikováno v:
2021 Innovations in Intelligent Systems and Applications Conference (ASYU).
In this paper, we improve the intent detection performance by leveraging the datasets which are collected in the same domain but have different intent categories. In the baseline system, we use the state-of-the-art pre-trained transformer models with
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::c0dde056f3114a32e78c36d9513404d3
https://doi.org/10.1109/asyu52992.2021.9599055
https://doi.org/10.1109/asyu52992.2021.9599055
Publikováno v:
SIU
The method frequently used for text classification is supervised modeling with a large training set with labels. In some cases, we may not have labeled data. Modeling in the absence of labeled data for target classes is called zero-shot modeling. For
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::5c81c202f2f125f381590045ed892e55
https://doi.org/10.1109/siu53274.2021.9477864
https://doi.org/10.1109/siu53274.2021.9477864
Autor:
Ömer Şayli, Levent M. Arslan
Publikováno v:
Dilbilim Araştırmaları Dergisi, Vol 14, Iss 1, Pp 15-26 (2003)
Durations of the Turkish phonemes are investigated in this study using the high quality digital records of two adult male utterances. Three recording types were used: reading of just one-word, reading of a phrase and reading of a sentence. The record
Externí odkaz:
https://doaj.org/article/71c7cc53119742749cc35fad42bac0b3
Autor:
Levent M. Arslan, Osman Büyük
Publikováno v:
Expert Systems. 38
Sequence to sequence models (seq2seq) require a large amount of labelled training data to learn the mapping between the input and output. A large set of misspelled words together with their corrections is needed to train a seq2seq spelling correction
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_dedup___::63aee19c78082dfa58e14f353018114a
https://doi.org/10.1111/exsy.12692
https://doi.org/10.1111/exsy.12692
Publikováno v:
Speech and Computer ISBN: 9783030878016
SPECOM
SPECOM
For the task of speaker recognition from audio, it is known that speakers experience different levels of error rates. In this work, predicting the proneness to false alarm and false reject of a given speaker embedding is investigated. Although exact
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::50fd0eeef9ec79f8358530218e96cea4
https://doi.org/10.1007/978-3-030-87802-3_74
https://doi.org/10.1007/978-3-030-87802-3_74
Publikováno v:
2020 28th Signal Processing and Communications Applications Conference (SIU).
Externí odkaz:
https://explore.openaire.eu/search/publication?articleId=doi_________::dac8fd19b1645c2bcc289ef4ec53187d
https://doi.org/10.1109/siu49456.2020.9302425
https://doi.org/10.1109/siu49456.2020.9302425