Towards a Deep Speech Model for Romanian Language

Panaite, Marilena; Ruşeţi, Ştefan; Dascălu, Mihai; Trăușan-Matu, Ștefan

doi:10.1109/cscs.2019.00076

Cited by 4 publications

(3 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While this does decrease the potential for robustness when developing a generalized speech recognition system, it is less of an issue when considering the integration of speech recognition into field work and documentation, as there tend to be few consultants providing large quantities of data each. This would also decrease our total quantity of data, but others have been successful using methods similar to those we outline above on smaller datasets (Meyer, 2019;Jimerson et al, 2018;Panaite et al, 2019;Yu et al, 2019).…”

Section: Discussionmentioning

confidence: 99%

Towards a Speech Recognizer for Komi, an Endangered and Low-Resource Uralic Language

Hjortnaes¹,

Partanen²,

Rießler³

et al. 2020

Proceedings of the Sixth International Workshop on Computational Linguistics of Uralic Languages

View full text Add to dashboard Cite

In this paper, we present and evaluate a first pass speech recognition model for Komi, an endangered and low-resource Uralic language spoken in Russia. We compare a transfer learning approach from English with a baseline model trained from scratch using DeepSpeech (an end-to-end ASR model) and evaluate the impact of fine tuning a language model for correcting the output of the network. We also provides an overview of previous research and perform an error analysis with a focus on the language model and the challenges introduced by a fieldwork based corpus. Though we only achieve a 70.9% Character Error Rate, there is a great deal to be learned from the circumstances presented by our data's structure and origins.

show abstract

Section: Discussionmentioning

confidence: 99%

Towards a Speech Recognizer for Komi, an Endangered and Low-Resource Uralic Language

Hjortnaes¹,

Partanen²,

Rießler³

et al. 2020

Proceedings of the Sixth International Workshop on Computational Linguistics of Uralic Languages

View full text Add to dashboard Cite

show abstract

“…It is most commonly used with large languages when the resources available vastly outnumber what we have. We found several other cases where DeepSpeech was used, for example, with Russian (Iakushkin et al, 2018), Romanian (Panaite et al, 2019), Tujian (Yu et al, 2019) and Bangla (Saurav et al, 2018). All of these experiments report higher scores than we do, with the exception of Russian, with smaller data, but there are important differences as well.…”

Section: Related Workmentioning

confidence: 95%

Proceedings of the Sixth International Workshop on Computational Linguistics of Uralic Languages

2020

View full text Add to dashboard Cite

This article introduces a novel and creative application of the Constraint Grammar formalism, by presenting an automated method for pseudonymising a Zyrian Komi spoken language corpus in an effective, reliable and scalable manner. The method is intended to be used to minimize various kinds of personal information found in the corpus in order to make spoken language data available while preventing the spread of sensitive personal data about the recorded informants or other persons mentioned in the texts. In our implementation, a Constraint Grammar based pseudonymisation tool is used as an automatically applied shallow layer that derives from the original corpus data a version which can be shared for open research use. TeesiqSeo artikli tutvustas vahtsõt ja loovat piirdmiisi grammatiga (PG) formalismõ pruuk´mist. Taas om metod´, kon PG pruugitas tuusjaos, et süräkomi kõnõldu keele korpusõ lindistuisi saassiq tegüsähe, kimmähe ja kontrol´misõvõimalusõga vaŕonimmiga käkkiq. Seo metod´om tett, et korpusõn saassiq kõnõlõjidõ andmit nii pall´o vähembäs võttaq, ku või, ja et tulõmit saassiq kergehe käsilde kontrolliq. Mi plaani perrä pruugitas taad ku automaatsõt kihti, miä tege korpusõ säändses, et taad või kergehe uuŕmisõ jaos jakaq.

show abstract

“…To facilitate the process of using it, a group of technologies appeared in multiple fields in DL. Several researchers tested the DeepSpeech model to build an SR system for Russian and German languages (Agarwal & Zesch, 2019;Panaite et al, 2019;Iakushkin et al, 2018).…”

Section: Background and Literature Reviewmentioning

confidence: 99%

Building a neural speech recognizer for quranic recitations

Al-Issa

Al‐Ayyoub

Al-Khaleel

et al. 2022

Int J Speech Technol

View full text Add to dashboard Cite

This work is an effort towards building Neural Speech Recognizers system for Quranic recitations that can be effectively used by anyone regardless of their gender and age. Despite having a lot of recitations available online, most of them are recorded by professional male adult reciters, which means that an ASR system trained on such datasets would not work for female/ child reciters. We address this gap by adopting a benchmark dataset of audio records of Quranic recitations that consists of recitations by both genders from different ages. Using this dataset, we build several speaker-independent NSR systems based on the DeepSpeech model and use word error rate (WER) for evaluating them. The goal is to show how an NSR system trained and tuned on a dataset of a certain gender would perform on a test set from the other gender. Unfortunately, the number of female recitations in our dataset is rather small while the number of male recitations is much larger. In the first set of experiments, we avoid the imbalance issue between the two genders and down-sample the male part to match the female part. For this small subset of our dataset, the results are interesting with 0.968 WER when the system is trained on male recitations and tested on female recitations. The same system gives 0.406 WER when tested on male recitations. On the other hand, training the system on female recitations and testing it on male recitation gives 0.966 WER while testing it on female recitations gives 0.608 WER.

show abstract

Towards a Deep Speech Model for Romanian Language

Cited by 4 publications

References 5 publications

Towards a Speech Recognizer for Komi, an Endangered and Low-Resource Uralic Language

Towards a Speech Recognizer for Komi, an Endangered and Low-Resource Uralic Language

Proceedings of the Sixth International Workshop on Computational Linguistics of Uralic Languages

Building a neural speech recognizer for quranic recitations

Contact Info

Product

Resources

About