Sorry, you need to enable JavaScript to visit this website.

Model based VAD approaches have been widely used and
achieved success in practice. These approaches usually cast
VAD as a frame-level classification problem and employ statistical
classifiers, such as Gaussian Mixture Model (GMM) or
Deep Neural Network (DNN) to assign a speech/silence label
for each frame. Due to the frame independent assumption classification,
the VAD results tend to be fragile. To address this
problem, in this paper, a new structured multi-frame prediction
DNN approach is proposed to improve the segment-level

To investigate the temporal dynamics and spatial representations in speech perception and processing, electroencephalograph (EEG) signals were recorded when subjects listening Chinese words and pseudo words.

This paper conducted an interface research on rhetorical and prosodic aspects of Chinese reading discourses within the Rhetorical Structure Theory (RST) framework. Ten discourses in 3 genres (Commentary, Narrative, and Descriptive) from the Annotated Speech Corpus of Chinese Discourse (ASCCD) were diagrammed in RST. The recordings from 5 males and 5 females were annotated and further analyzed acoustically and statistically by applying Praat and R.

We propose a novel speaker-dependent (SD) approach to joint training of deep neural networks (DNNs) with an explicit speech separation structure for multi-talker speech recognition in a single-channel setting. First, a multi-condition training strategy is designed for a SD-DNN recognizer in multi-talker scenarios, which can significantly reduce the decoding runtime and improve the recognition accuracy over the approaches that use speaker-independent DNN models with a complicated joint decoding framework.

This study presents an approach to dialog state tracking (DST) in an interview conversation by using the long short-term memory (LSTM) and artificial neural network (ANN). First, the techniques of word embedding are employed for word representation by using the word2vec model. Then, each input sentence is represented by a sentence hidden vector using the LSTM-based sentence model. The sentence hidden vectors for each sentence are fed to the LSTM-based answer model to map the interviewee’s answer to an answer hidden vector.

This study presents an approach to dialog state tracking (DST) in an interview conversation by using the long short-term memory (LSTM) and artificial neural network (ANN). First, the techniques of word embedding are employed for word representation by using the word2vec model. Then, each input sentence is represented by a sentence hidden vector using the LSTM-based sentence model. The sentence hidden vectors for each sentence are fed to the LSTM-based answer model to map the interviewee’s answer to an answer hidden vector.

The widely adopted i-vector performances well in text-independent speaker verification with long speech duration. How to integrate the state-of-the-art i-vector framework into the text-prompted speaker verification is addressed in this paper. To take advantage of the lexical information and enhance the performance for speaker verification with random digit sequences, this paper proposes to extract a set of digit-dependent local i-vectors from the utterance instead of extracting a single i-vector. The digit-dependent local i-vector is considered

The paper proposed a method to realize a speech-to-gesture conversion for communication between normal and speech-impaired people. Keyword spotting was employed to recognize the keywords from input speech signals. At the same time, the three dimensional gesture models of keywords were built by 3D modeling technology according to the "Chinese sign language". The speech-to-gesture conversion was finally realized by playing the corresponding 3D gestures with OpenGL from the results of keyword spotting.

Natural and synthesized speech in L2 Mandarin produced by American English learners was evaluated by native Mandarin speakers to identify focus status and rate the naturalness of the speech. The results reveal that natural speech was recognized and rated better than synthesized speech, early learners’ speech better than late learners’ speech, focused sentences better than no-focus sentences, and initial focus and medial focus better than final focus. Tones of in-focus words interacted with focus status of the sentence and speaker group.