Wang, Jun

Permanent URI for this collectionhttps://hdl.handle.net/10735.1/4986

Dr. Jun Wang is an Assistant Professor of Bioengineering and Principle Investigator of the Speech Disorders and Technology Lab. He conducts both science and engineering research in speech disorders. Topis include:

  • Silent speech recognition / interface
  • Articulation-to-speech synthesis
  • Dysarthric speech recognition
  • Motor speech disorders due to amyotrophic lateral sclerosis
  • Articulatory patterns after laryngectomy

ORCID page

Browse

Recent Submissions

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Automatic Speech Activity Recognition from MEG Signals Using Seq2Seq Learning
    (IEEE Computer Society, 2019-03) Dash, Debadatta; Ferrari, P.; Malik, S.; Wang, Jun; 0000-0001-7265-217X (Wang, J); Dash, Debadatta; Wang, Jun
    Accurate interpretation of speech activity from brain signals is critical for understanding the relationship between neural patterns and speech production. Current research on speech activity recognition from the brain activity heavily relies on the region of interest (ROI) based functional connectivity analysis or source separation strategies to map the activity as a spatial localization of a brain function. Albeit effective, these methods require prior knowledge of the brain and expensive computational effort. In this study, we investigated automatic speech activity recognition from brain signals using machine learning. Neural signals of four subjects during four stages of a speech task (i.e., rest, perception, preparation, and production) were recorded using magnetoencephalography (MEG), which has an excellent temporal and spatial resolution. First, a deep neural network (DNN) was used to classify the four whole tasks from the MEG signals. Further, we trained a sequence to sequence (Seq2Seq) long short-term memory-recurrent neural network (LSTM-RNN) for continuous (sample by sample) prediction of the speech stages/tasks by leveraging its sequential pattern learning paradigm. Experimental results indicate the effectiveness of both DNN and LSTM-RNN for automatic speech activity recognition from MEG signals. © 2019 IEEE.
  • Thumbnail Image
    Item
    Using Electromagnetic Articulography with a Tongue Lateral Sensor to Discriminate Manner of Articulation
    (Acoustical Society of America) Katz, William F.; Mehta, Sonya; Wood, Matthew; Wang, Jun; Katz, William F.; Mehta, Sonya; Wood, Matthew; Wang, Jun
    This study examined the contributions of the tongue tip (TT), tongue body (TB), and tongue lateral (TL) sensors in the electromagnetic articulography (EMA) measurement of American English alveolar consonants. Thirteen adults produced /ɹ/, /l/, /z/ and /d/ in /αCα/ syllables while being recorded with an EMA system. According to statistical analysis of sensor movement and the results of a machine classification experiment, the TT sensor contributed most to consonant differences, followed by TB. The TL sensor played a complementary role, particularly for distinguishing /z/. © 2017 Acoustical Society of America.
  • Thumbnail Image
    Item
    Parkinson's Condition Estimation Using Speech Acoustic and Inversely Mapped Articulatory Data
    (International Speech and Communication Association) Hahm, Seongjun; Wang, Jun; Noth E.; Steidl S.; Moller S.; Ney H.; Mobius B.; Hahm, Seongjun; Wang, Jun
    Parkinson's disease is a neurological disorder that affects patient's motor function including speech articulation. There is no cure for Parkinson's disease. Speech and motor function declines as the disease progresses. Automatic assessment of the disease condition may advance the treatment of Parkinson's disease with objective, inexpensive measures. Speech acoustics, which can be easily obtained from patients, has been used for automatic assessment. The use of information in motor function of articulator (e.g., jaw, tongue, or lips) has rarely been investigated. In this paper, we proposed an approach of automatic assessment of Parkinson's condition using both acoustic data and acoustically-inverted articulatory data. The quasi-articulatory features were obtained from the Parkinson's acoustic speech data using acoustic-to-articulatory inverse mapping. Support vector regression (SVR) and deep neural network (DNN) regression were used in the experiment. Results indicated adding articulatory data to acoustic data can improve the performance of using acoustic data only, for both SVR and DNN. In addition, deep neural network outperformed support vector regression on the same data features measured with Pearson correlation but not with Spearman correlation. The implications of our approach with further improvement were discussed.

Works in Treasures @ UT Dallas are made available exclusively for educational purposes such as research or instruction. Literary rights, including copyright for published works held by the creator(s) or their heirs, or other third parties may apply. All rights are reserved unless otherwise indicated by the copyright owner(s).