Improving Low Resource Turkish Speech Recognition with Data Augmentation and TTS

Gokay, Ramazan; Yalçın, Hülya

doi:10.1109/ssd.2019.8893184

Cited by 26 publications

(16 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Selftraining, or sequnece-level knowledge distillation by text-to-text machine translation model, is the most effective way to utilize the huge ASR training data . On the other hand, synthesizing data by text-to-speech (TTS) has been demonstrated to be effective for low resource speech recognition task (Gokay and Yalcin, 2019;Ren et al, 2019). To the best of our knowledge, this is the first work to augment data by TTS for simultaneous speech-to-text translation tasks.…”

Section: Related Workmentioning

confidence: 99%

The USTC-NELSLIP Systems for Simultaneous Speech Translation Task at IWSLT 2021

Liú¹,

Du²,

Li³

et al. 2021

Proceedings of the 18th International Conference on Spoken Language Translation (IWSLT 2021)

View full text Add to dashboard Cite

This paper describes USTC-NELSLIP's submissions to the IWSLT2021 Simultaneous Speech Translation task. We proposed a novel simultaneous translation model, Cross Attention Augmented Transducer (CAAT), which extends conventional RNN-T to sequence-tosequence tasks without monotonic constraints, e.g., simultaneous translation. Experiments on speech-to-text (S2T) and text-to-text (T2T) simultaneous translation tasks shows CAAT achieves better quality-latency trade-offs compared to wait-k, one of the previous state-ofthe-art approaches. Based on CAAT architecture and data augmentation, we build S2T and T2T simultaneous translation systems in this evaluation campaign. Compared to last year's optimal systems, our S2T simultaneous translation system improves by an average of 11.3 BLEU for all latency regimes, and our T2T simultaneous translation system improves by an average of 4.6 BLEU.

show abstract

Section: Related Workmentioning

confidence: 99%

The USTC-NELSLIP Systems for Simultaneous Speech Translation Task at IWSLT 2021

Liú¹,

Du²,

Li³

et al. 2021

Proceedings of the 18th International Conference on Spoken Language Translation (IWSLT 2021)

View full text Add to dashboard Cite

show abstract

“…Data from other languages were additionally used in [64]. Acoustic data perturbation and speech synthesis were combined in [65], resulting in a 14.8% relative WER improvement. Semisupervised training refers to train the model with supervised data and unsupervised data through the confidence threshold.…”

Section: A．data Augmentationmentioning

confidence: 99%

“…Due to the change of the signal length, the GMM-HMM system was used to realign the data after the speed perturbation [70]. Gokay et al [65] used speed perturbation, volume perturbation and a combination of the two for data augmentation. Kanda et al [72] studied three distortion methods of vocal tract length distortion, speech rate distortion, and frequency-axis random distortion.…”

Section: ) Acoustic Data Perturbationmentioning

confidence: 99%

See 1 more Smart Citation

Acoustic Modeling Based on Deep Learning for Low-Resource Speech Recognition: An Overview

Kang

Chen

et al. 2020

IEEE Access

View full text Add to dashboard Cite

The polarization of world languages is becoming more and more obvious. Many languages, mainly endangered languages, are of low-resource attribute due to lack of information. Both language conservation and cultural heritage face important challenges. Therefore, speech recognition for lowresource scenario has become a hot topic in the field of speech. Based on the complex network structures and huge model parameters, deep learning has become a powerful science in the process of speech recognition, which has a broad and far-reaching significance for the study of low-resource speech recognition. Aiming at the characteristic of low resource, this paper reviews the history and research status of two kinds of acoustic models of deep learning neural networks and acoustic end-to-end structures. We further elaborate on several key techniques for improving performance in the two aspects of data and model training. There are two projects for low-resource languages introduced in this paper. The possible future developments are finally pointed out. These works provide some reference for computer speech and language processing.

show abstract

“…In understanding grammatical ambiguous sentences, the system requires different time to process each sentence; the processing of this sentence depends on the number of characters understood [34]. The average sentence search value is 0.003275.…”

Section: The Speed In Detecting Ambiguous Wordsmentioning

confidence: 99%

Detecting Indonesian ambiguous sentences using Boyer-Moore algorithm

et al. 2020

View full text Add to dashboard Cite

Ambiguous sentences are divided into 3 types namely phonetic, lexical, and grammatical. This study focuses on grammatical ambiguous sentences, grammatical ambiguous sentences are ambiguities that occur due to incorrect grammar, but this ambiguity will disappear once it is used within a sentence. Ambiguous sentences become a big problem when they are processed by a computer. In order for the computer to interpret ambiguous words correctly, this study seeks to develop detection of Indonesian ammbiguous sentences using Boyer Moore algorithm. This algorithm matches ambiguous sentences that are inserted as input with the data set. Then the sentence is being detected whether it contains ambiguous sentences, by calculating the percentage of similarity using cosine similarity method. Cosine similarity system is able to find out the meaning of the sentence. In the data set, the number of ambiguous sentences that can be collected is 50 words. The 50 words consist of ambiguous words data, ambiguous sentences, and ambiguous sentence meanings. This system trial was carried out for 200 times and the accuracy level was 0.935, precision was 0.9320, and Recall was 0.8. While the F-Measure was 0.8061. While the speed for word search 0.003275 seconds

show abstract

Improving Low Resource Turkish Speech Recognition with Data Augmentation and TTS

Cited by 26 publications

References 5 publications

The USTC-NELSLIP Systems for Simultaneous Speech Translation Task at IWSLT 2021

The USTC-NELSLIP Systems for Simultaneous Speech Translation Task at IWSLT 2021

Acoustic Modeling Based on Deep Learning for Low-Resource Speech Recognition: An Overview

Detecting Indonesian ambiguous sentences using Boyer-Moore algorithm

Contact Info

Product

Resources

About