Low-Resource Expressive Text-To-Speech Using Data Augmentation

AbstractDeep learning techniques are currently being applied in automated text-to-speech (TTS) systems, resulting in significant improvements in performance. However, these methods require large amounts of text-speech paired data for model training, and collecting this data is costly. Therefore, in this paper, we propose a single-speaker TTS system containing both a spectrogram prediction network and a neural vocoder for the target language, using only 30 min of target language text-speech paired data for training. We evaluate three approaches for training the spectrogram prediction models of our TTS system, which produce mel-spectrograms from the input phoneme sequence: (1) cross-lingual transfer learning, (2) data augmentation, and (3) a combination of the previous two methods. In the cross-lingual transfer learning method, we used two high-resource language datasets, English (24 h) and Japanese (10 h). We also used 30 min of target language data for training in all three approaches, and for generating the augmented data used for training in methods 2 and 3. We found that using both cross-lingual transfer learning and augmented data during training resulted in the most natural synthesized target speech output. We also compare single-speaker and multi-speaker training methods, using sequential and simultaneous training, respectively. The multi-speaker models were found to be more effective for constructing a single-speaker, low-resource TTS model. In addition, we trained two Parallel WaveGAN (PWG) neural vocoders, one using 13 h of our augmented data with 30 min of target language data and one using the entire 12 h of the original target language dataset. Our subjective AB preference test indicated that the neural vocoder trained with augmented data achieved almost the same perceived speech quality as the vocoder trained with the entire target language dataset. Overall, we found that our proposed TTS system consisting of a spectrogram prediction network and a PWG neural vocoder was able to achieve reasonable performance using only 30 min of target language training data. We also found that by using 3 h of target language data, for training the model and for generating augmented data, our proposed TTS model was able to achieve performance very similar to that of the baseline model, which was trained with 12 h of target language data.

Download Full-text

Support vector approach using data augmentation with copula function applied to load forecasting

10.26678/abcm.cobem2019.cob2019-1042 ◽

2019 ◽

Author(s):

Gabriel Ribeiro ◽

Marcos Yamasaki ◽

Helon Vicente Hultmann Ayala ◽

Leandro Coelho ◽

Viviana Mariani

Keyword(s):

Data Augmentation ◽

Load Forecasting ◽

Copula Function ◽

Support Vector ◽

Using Data ◽

Vector Approach

Download Full-text

Implementation of Concatenation Technique for Low Resource Text-To-Speech System Based on Marathi Talking Calculator

10.21437/sltu.2018-16 ◽

2018 ◽

Author(s):

Monica Mundada ◽

Sangramsing Kayte ◽

Pradip Das

Keyword(s):

Text To Speech ◽

Low Resource

Download Full-text

UAV Image Multi-Labeling with Data-Efficient Transformers

Applied Sciences ◽

10.3390/app11093974 ◽

2021 ◽

Vol 11 (9) ◽

pp. 3974

Author(s):

Laila Bashmal ◽

Yakoub Bazi ◽

Mohamad Mahmoud Al Rahhal ◽

Haikel Alhichri ◽

Naif Al Ajlan

Keyword(s):

Data Augmentation ◽

Feature Representation ◽

Aerial Image ◽

Remote Sensing Images ◽

Training Set ◽

Proposed Model ◽

Class Labels ◽

Using Data ◽

Uav Image

In this paper, we present an approach for the multi-label classification of remote sensing images based on data-efficient transformers. During the training phase, we generated a second view for each image from the training set using data augmentation. Then, both the image and its augmented version were reshaped into a sequence of flattened patches and then fed to the transformer encoder. The latter extracts a compact feature representation from each image with the help of a self-attention mechanism, which can handle the global dependencies between different regions of the high-resolution aerial image. On the top of the encoder, we mounted two classifiers, a token and a distiller classifier. During training, we minimized a global loss consisting of two terms, each corresponding to one of the two classifiers. In the test phase, we considered the average of the two classifiers as the final class labels. Experiments on two datasets acquired over the cities of Trento and Civezzano with a ground resolution of two-centimeter demonstrated the effectiveness of the proposed model.

Download Full-text

Exploring the octanol–water partition coefficient dataset using deep learning techniques and data augmentation

Communications Chemistry ◽

10.1038/s42004-021-00528-9 ◽

2021 ◽

Vol 4 (1) ◽

Author(s):

Nadin Ulrich ◽

Kai-Uwe Goss ◽

Andrea Ebert

Keyword(s):

Partition Coefficient ◽

Data Augmentation ◽

Environmental Chemistry ◽

Chemical Properties ◽

Predictive Performance ◽

Chemical Structures ◽

Tautomeric Forms ◽

Learning Techniques ◽

Water Partition Coefficient ◽

Using Data

AbstractToday more and more data are freely available. Based on these big datasets deep neural networks (DNNs) rapidly gain relevance in computational chemistry. Here, we explore the potential of DNNs to predict chemical properties from chemical structures. We have selected the octanol-water partition coefficient (log P) as an example, which plays an essential role in environmental chemistry and toxicology but also in chemical analysis. The predictive performance of the developed DNN is good with an rmse of 0.47 log units in the test dataset and an rmse of 0.33 for an external dataset from the SAMPL6 challenge. To this end, we trained the DNN using data augmentation considering all potential tautomeric forms of the chemicals. We further demonstrate how DNN models can help in the curation of the log P dataset by identifying potential errors, and address limitations of the dataset itself.

Download Full-text

Minimization of CNN Training Data by using Data Augmentation for Inline Defect Classification

2020 International Symposium on Semiconductor Manufacturing (ISSM) ◽

10.1109/issm51728.2020.9377504 ◽

2020 ◽

Author(s):

Akihiro Fujishiro ◽

Yoshikazu Nagamura ◽

Tatsuya Usami ◽

Masao Inoue

Keyword(s):

Data Augmentation ◽

Training Data ◽

Defect Classification ◽

Using Data

Download Full-text

Improved neural machine translation for low-resource English–Assamese pair

Journal of Intelligent & Fuzzy Systems ◽

10.3233/jifs-219260 ◽

2021 ◽

pp. 1-12

Author(s):

Sahinur Rahman Laskar ◽

Abdullah Faiz Ur Rahman Khilji ◽

Partha Pakray ◽

Sivaji Bandyopadhyay

Keyword(s):

Machine Translation ◽

Data Augmentation ◽

Language Translation ◽

Linguistically Diverse ◽

Neural Machine Translation ◽

Low Resource ◽

Parallel Data ◽

The World ◽

Translation Accuracy ◽

Vocabulary Problems

Language translation is essential to bring the world closer and plays a significant part in building a community among people of different linguistic backgrounds. Machine translation dramatically helps in removing the language barrier and allows easier communication among linguistically diverse communities. Due to the unavailability of resources, major languages of the world are accounted as low-resource languages. This leads to a challenging task of automating translation among various such languages to benefit indigenous speakers. This article investigates neural machine translation for the English–Assamese resource-poor language pair by tackling insufficient data and out-of-vocabulary problems. We have also proposed an approach of data augmentation-based NMT, which exploits synthetic parallel data and shows significantly improved translation accuracy for English-to-Assamese and Assamese-to-English translation and obtained state-of-the-art results.

Download Full-text

Text Mining Drug-Protein Interactions using an Ensemble of BERT, Sentence BERT and T5 models

10.1101/2021.10.26.465944 ◽

2021 ◽

Author(s):

Xin Sui ◽

Wanjing Wang ◽

Jinfeng Zhang

Keyword(s):

Protein Interactions ◽

Clustering Algorithm ◽

Data Augmentation ◽

Majority Vote ◽

Classification Model ◽

Ensemble Model ◽

K Nearest Neighbors ◽

Test Dataset ◽

Improved Performance ◽

Using Data

In this work, we trained an ensemble model for predicting drug-protein interactions within a sentence based on only its semantics. Our ensembled model was built using three separate models: 1) a classification model using a fine-tuned BERT model; 2) a fine-tuned sentence BERT model that embeds every sentence into a vector; and 3) another classification model using a fine-tuned T5 model. In all models, we further improved performance using data augmentation. For model 2, we predicted the label of a sentence using k-nearest neighbors with its embedded vector. We also explored ways to ensemble these 3 models: a) we used the majority vote method to ensemble these 3 models; and b) based on the HDBSCAN clustering algorithm, we trained another ensemble model using features from all the models to make decisions. Our best model achieved an F-1 score of 0.753 on the BioCreative VII Track 1 test dataset.

Download Full-text