Compositional Neural Network Language Models for Agglutinative Languages

Arisoy, Ebru; Saraçlar, Murat

doi:10.21437/interspeech.2016-1239

Cited by 7 publications

(3 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Özellikle uzun ve zamana bağlı bilgilerin modellemesindeki zorlukları çözebilmek için Uzun Kısa Süreli Bellek (LSTM: Long Short Term Memory) Hochreiter ve Schmidhuber tarafından önerilmiştir [25]. Arısoy ve Saraçlar LSTM yapısını Türkçe yayın haberlerinin transkripsiyonu için kullanmış ve tekrarlayan sinir ağlarını kullanarak bir ASR sistemi geliştirmiştir [26].…”

Section: Literatür Taraması (Literature Review)unclassified

Otomatik Konuşma Tanımaya Genel Bakış, Yaklaşımlar ve Zorluklar: Türkçe Konuşma Tanımanın Gelecekteki Yolu

Oyucu

Sever

Polat

2019

Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım Ve Teknoloji

View full text Add to dashboard Cite

Figure A shows the application area of the Applications Automatic Speech Recognition (ASR) system. Adaptation is required for adaptation processes to be carried out according to the Application area. During the Speech Processing phase, which is the first entry point of the ASR system, feature extraction is performed from the audio signal. Individual properties are obtained by different feature extraction techniques. For example, Mel Frequency Cepstral Coefficient (MFCC) is a feature extraction technique commonly used in speech recognition systems. Decoder, one of the other components of ASR, converts the feature vectors obtained by using Acoustic Model (AM) and Language Model (LM) into phoneme sequences. In acoustic modeling, firstly, the posterior probability of the phoneme within a given time signal is calculated. In the artificial neural network-based acoustic model, the posterior probability of phonemes is independent for each window. This independence means that the phonemes in a word are independent of each other. Figure A. Basic architecture of speech recognition system.Purpose: This study presents a literature review on speech recognition and then discusses the recorded signs of progress made in this area of research for different languages. The data sets used in speech recognition systems, feature extraction approaches, speech recognition methods and performance evaluation criteria are examined and the focus is on the development of speech recognition and the difficulties in this field. Theory and Methods:In this study, literature review (systematic), which is an important component for a scientific article, was carried out. This process was carried out by the combination of different methods. A combination of review approaches is given.Results: According to the information obtained as a result of the research; Computational architectures that can be applied to resistance to the acoustic environment, self-learning in ASR, detection of unknown words, the success of the Turkish ASR at a broad and limited repertoire level, insufficient source status and Automatic Speech Recognition ASR were evaluated. In addition, the future of Turkish ASR was discussed and recommendations were made to overcome the current difficulties for Turkish ASR. Conclusion:The aim of this study is to examine the current speech recognition methods and approaches and to present the developments in this field in detail. For this reason, approaches, datasets and the difficulties faced by the researchers in their studies in this field are discussed in the scope of the study. The effect of deep learning and classical approaches on ASR was investigated. A road map is provided for researchers to incorporate the detailed information necessary for their work in this field to their own work and to overcome the present challenges.

show abstract

Section: Literatür Taraması (Literature Review)unclassified

Otomatik Konuşma Tanımaya Genel Bakış, Yaklaşımlar ve Zorluklar: Türkçe Konuşma Tanımanın Gelecekteki Yolu

Oyucu

Sever

Polat

2019

Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım Ve Teknoloji

View full text Add to dashboard Cite

show abstract

“…The purposed system is trained by Turkish News TV program recordings and Law lecture video recordings. In [14], DNN is also used for building acoustic model. In [15], GMM and DNN based models are trained and tested using the corpus which is developed in [16].…”

Section: Introductionmentioning

confidence: 99%

Deep Learning Based Automatic Speech Recognition for Turkish

Tombaloğlu¹,

Erdem

2020

Sakarya University Journal of Science

View full text Add to dashboard Cite

Using Deep Neural Networks (DNN) as an advanced Artificial Neural Networks (ANN) has become widespread with the development of computer technology. Although DNN has been applied for solving Automatic Speech Recognition (ASR) problem in some languages, DNNbased Turkish Speech Recognition has not been studied extensively. Turkish language is an agglutinative and a phoneme-based language. In this study, a Deep Belief Network (DBN) based Turkish phoneme and speech recognizer is developed. The proposed system recognizes words in the system vocabulary and phoneme components of out of vocabulary (OOV) words. Sub-word (morpheme) based language modelling is implemented into the system. Each phoneme of Turkish language is also modelled as a sub-word in the model. Sub-word (morpheme) based language model is widely used for agglutinative languages to prevent excessive vocabulary size. The performance of the suggested DBN based ASR system is compared with the conventional recognition method, GMM (Gaussian Mixture Method) based Hidden Markov Model (HMM). Regarding to performance metrics, the recognition rate of Turkish language is improved in compare with previous studies.

show abstract

“…In [17], vowel harmony rule in Turkish is incorporated to SR. In [18], long shortterm memory (LSTM) and compositional LSTM recurrent neural networks are used for Turkish broadcast news transcription task. In [19], a speaker independent and phone based continuous time digit sequence recognition system was designed.…”

Section: Introductionmentioning

confidence: 99%

Turkish Speech Recognition Based On Deep Neural Networks

Kimanuka¹,

Büyük²

2018

SDÜ Fen Bil Enst Der

View full text Add to dashboard Cite

In this paper we develop a Turkish speech recognition (SR) system using deep neural networks and compare it with the previous state-of-the-art traditional Gaussian mixture model-hidden Markov model (GMM-HMM) method using the same Turkish speech dataset and the same large vocabulary Turkish corpus. Nowadays most SR systems deployed worldwide and particularly in Turkey use Hidden Markov Models to deal with the speech temporal variations. Gaussian mixture models are used to estimate the amount at which each state of each HMM fits a short frame of coefficients which is the representation of an acoustic input. A deep neural network consisting of feed-forward neural network is another way to estimate the fit; this neural network takes as input several frames of coefficients and gives as output posterior probabilities over HMM states. It has been shown that the use of deep neural networks can outperform the traditional GMM-HMM in other languages such as English and German. The fact that Turkish language is an agglutinative language and the lack of a huge amount of speech data complicate the design of a performant SR system. By making use of deep neural networks we will obviously improve the performance but still we will not achieve better result than English language due to the difference in the availability of speech data. We present various architectural and training techniques for the Turkish DNN-based models. The models are tested using a Turkish database collected from mobile devices. In the experiments, we observe that the Turkish DNN-HMM system have decreased the word error rate approximately 2.5% when compared to the GMM-HMM traditional system.

show abstract

Compositional Neural Network Language Models for Agglutinative Languages

Cited by 7 publications

References 20 publications

Otomatik Konuşma Tanımaya Genel Bakış, Yaklaşımlar ve Zorluklar: Türkçe Konuşma Tanımanın Gelecekteki Yolu

Otomatik Konuşma Tanımaya Genel Bakış, Yaklaşımlar ve Zorluklar: Türkçe Konuşma Tanımanın Gelecekteki Yolu

Deep Learning Based Automatic Speech Recognition for Turkish

Turkish Speech Recognition Based On Deep Neural Networks

Contact Info

Product

Resources

About