DirecTL

Jiampojamarn, Sittichai; Bhargava, Aditya; Dou, Qing; Dwyer, Kenneth; Kondrak, Grzegorz

doi:10.3115/1699705.1699712

Cited by 20 publications

(4 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here, g represents a given grapheme sequence, where ϕ is the most likely pronunciation of the grapheme sequence g. Soon after, Jiampojamarn et al represented the joint n-grams model for G2P conversion as an online discriminative sequence-prediction model, which used a many-tomany alignment between grapheme and phoneme sequences and a feature vector consisting of n-grams context features, HMM-like transition features, and linear-chain features [6]. For each training iteration, the feature weight vector was updated using the MIRA algorithm; this system is called DirecTL.…”

Section: Mira-based Methods For G2p Conversion (Directl+)mentioning

confidence: 99%

“…Thus, researchers have proposed various data-driven meth-ods using many-to-many mapping techniques between graphemes and phonemes. Methods have been proposed based on hidden Markov models (HMMs) [2], [3], artificial neural network (ANNs) [4], joint-sequences [5], margininfused relaxed algorithms (MIRAs) [6], [7], a weighted finite-state transducer (WFST) [8], an adaptive regularization of weight vectors (AROW) [9], a narrow adaptive regularization of weight vectors (NAROW) [10], and structured soft-margin confidence weighted learning (SSMCW) [11]. Most of these methods, and especially SSMCW-based G2P conversion that is implemented in the Slearp toolkit * , have demonstrated significantly accurate results.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Using Reversed Sequences and Grapheme Generation Rules to Extend the Feasibility of a Phoneme Transition Network-Based Grapheme-to-Phoneme Conversion

Kheang

Katsurada

Iribe

et al. 2016

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

The automatic transcription of out-of-vocabulary words into their corresponding phoneme strings has been widely adopted for speech synthesis and spoken-term detection systems. By combining various methods in order to meet the challenges of grapheme-to-phoneme (G2P) conversion, this paper proposes a phoneme transition network (PTN)-based architecture for G2P conversion. The proposed method first builds a confusion network using multiple phoneme-sequence hypotheses generated from several G2P methods. It then determines the best finaloutput phoneme from each block of phonemes in the generated network. Moreover, in order to extend the feasibility and improve the performance of the proposed PTN-based model, we introduce a novel use of right-to-left (reversed) grapheme-phoneme sequences along with grapheme-generation rules. Both techniques are helpful not only for minimizing the number of required methods or source models in the proposed architecture but also for increasing the number of phoneme-sequence hypotheses, without increasing the number of methods. Therefore, the techniques serve to minimize the risk from combining accurate and inaccurate methods that can readily decrease the performance of phoneme prediction. Evaluation results using various pronunciation dictionaries show that the proposed model, when trained using the reversed grapheme-phoneme sequences, often outperformed conventional left-to-right grapheme-phoneme sequences. In addition, the evaluation demonstrates that the proposed PTN-based method for G2P conversion is more accurate than all baseline approaches that were tested.

show abstract

Section: Mira-based Methods For G2p Conversion (Directl+)mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Using Reversed Sequences and Grapheme Generation Rules to Extend the Feasibility of a Phoneme Transition Network-Based Grapheme-to-Phoneme Conversion

Kheang

Katsurada

Iribe

et al. 2016

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

show abstract

“…Once these mappings are learned, a common approach involves using a generative model that attempts to generate all possible transliterations of a source word, given the character mappings between two languages, and restricting the output to words in the target language (El-Kahki et al, 2011;Noeman and Madkour, 2010). Other approaches include the use of locality sensitive hashing (Udupa and Kumar, 2010) and classification (Jiampojamarn et al, 2010). Another dramatically different approaches involves the unsupervised learning of transliteration mappings from a large parallel corpus instead of transliteration pairs (Sajjad et al, 2012).…”

Section: Discussionmentioning

confidence: 99%

Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)

2014

View full text Add to dashboard Cite

There has been a lot of progress in the last 15 years in the area of Arabic Natural Language Processing (NLP). In particular, the TIDES, GALE, and BOLT programs provided a significant boost to Arabic NLP, both in generating new language and speech resources on a large scale, and in advancing the stateof-the-art in morphological processing, parsing, named entity recognition, information retrieval, speech recognition, and machine translation. The substantial investment done through these projects reflect the fact that the Middle East continues to grow in its political and economic importance. We also observe that countries in the Middle East invest substantially into higher education and into building an ecosystem, which fosters new research initiatives. This creates the hope that our own research field, NLP, and especially Arabic NLP will continue to grow, will continue to attract students both in the region and in top international universities, and that new job opportunities will open up not only in the well established language service providers, but also through start-ups offering innovative solutions.A number of Arabic NLP (or Arabic NLP-related) workshops and conferences have taken place, both in the Arab World and in association with international conferences. The Arabic NLP workshop at EMNLP 2014 follows in the footsteps of these previous efforts to provide a forum for researchers to share and discuss their ongoing work. The Arabic NLP workshop also includes a shared task on Automatic Arabic Error Correction, which was designed in the tradition of high profile NLP shared tasks such as CONLL's grammar/error detection and correction shared tasks in 2011-2013, and numerous machine translation campaigns by NIST/WMT/MEDAR, among others. The challenge chosen for the shared task is highly relevant, not only to spelling correction while composing a text, but also to developing techniques for automatically correcting errors in the far-from-perfect outputs of NLP technologies such as speech recognition or machine translation.We are happy to have received 40 submissions. Unfortunately, not all the papers could be included in the workshop due to time limitations. The acceptance rate is 50%. The papers cover a wide range of topics: building language resources for standard and dialectal Arabic, language identification, sentiment analysis, named entity disambiguation, and machine translation for dialectal Arabic, etc. Twelve papers were selected for oral presentation and were organized under the general topics Corpora (four papers), Text Mining (four papers), Translation & Transliteration (three papers) and one paper describing the shared task. The remaining eight papers were selected to be presented in a poster session. There is no difference in quality between the oral and poster presentations.The shared task was a success. We received 18 systems submissions from nine teams in six countries, representing a diverse set of approaches. Nine shared task system description (short) papers are included in the proceedings to ...

show abstract

“…The typical first step in training a G2P system involves aligning the corresponding grapheme and phoneme sequences in the input training dictionary. The approach adopted in this work is based on the EM driven multiple-to-multiple alignment algorithm proposed in Jiampojamarn et al (2007) and extended in Jiampojamarn and Kondrak (2010). This algorithm is capable of learning complex G↔P relationships like → //, and represents an improvement over earlier one-to-one stochastic alignment algorithms such as that introduced in Yianilos and Ristad (1998).…”

Section: Grapheme-to-phoneme Alignmentmentioning

confidence: 99%

Phonetisaurus: Exploring grapheme-to-phoneme conversion with joint n-gram models in the WFST framework

2015

View full text Add to dashboard Cite

This paper provides an analysis of several practical issues related to the theory and implementation of Grapheme-to-Phoneme (G2P) conversion systems utilizing the Weighted Finite-State Transducer paradigm. The paper addresses issues related to system accuracy, training time and practical implementation. The focus is on joint n-gram models which have proven to provide an excellent trade-off between system accuracy and training complexity. The paper argues in favor of simple, productive approaches to G2P, which favor a balance between training time, accuracy and model complexity. The paper also introduces the first instance of using joint sequence RnnLMs directly for G2P conversion, and achieves new state-of-the-art performance via ensemble methods combining RnnLMs and n-gram based models. In addition to detailed descriptions of the approach, minor yet novel implementation solutions, and experimental results, the paper introduces Phonetisaurus, a fully-functional, flexible, open-source, BSD-licensed G2P conversion toolkit, which leverages the OpenFst library. The work is intended to be accessible to a broad range of readers.

show abstract

DirecTL

Cited by 20 publications

References 7 publications

Using Reversed Sequences and Grapheme Generation Rules to Extend the Feasibility of a Phoneme Transition Network-Based Grapheme-to-Phoneme Conversion

Using Reversed Sequences and Grapheme Generation Rules to Extend the Feasibility of a Phoneme Transition Network-Based Grapheme-to-Phoneme Conversion

Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)

Phonetisaurus: Exploring grapheme-to-phoneme conversion with joint n-gram models in the WFST framework

Contact Info

Product

Resources

About