Double articulation analyzer with deep sparse autoencoder for unsupervised word discovery from speech signals

Taniguchi, Tadahiro; Nakashima, Ryoichi; Liu, Hailong; Nagasaka, Shogo

doi:10.1080/01691864.2016.1159981

Cited by 27 publications

(26 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, we briefly introduce NPB-DAA (Taniguchi et al, 2016b). Secondly, we describe DSAE-PBHL after introducing DSAE (Ng, 2011a;Liu et al, 2015;Taniguchi et al, 2016c).…”

Section: Methodsmentioning

confidence: 99%

Unsupervised Phoneme and Word Discovery From Multiple Speakers Using Double Articulation Analyzer and Neural Network With Parametric Bias

2019

Self Cite

View full text Add to dashboard Cite

This paper describes a new unsupervised machine learning method for simultaneous phoneme and word discovery from multiple speakers. Human infants can acquire knowledge of phonemes and words from interactions with his/her mother as well as with others surrounding him/her. From a computational perspective, phoneme and word discovery from multiple speakers is a more challenging problem than that from one speaker because the speech signals from different speakers exhibit different acoustic features. This paper proposes an unsupervised phoneme and word discovery method that simultaneously uses nonparametric Bayesian double articulation analyzer (NPB-DAA) and deep sparse autoencoder with parametric bias in hidden layer (DSAE-PBHL). We assume that an infant can recognize and distinguish speakers based on certain other features, e.g., visual face recognition. DSAE-PBHL is aimed to be able to subtract speaker-dependent acoustic features and extract speaker-independent features. An experiment demonstrated that DSAE-PBHL can subtract distributed representations of acoustic signals, enabling extraction based on the types of phonemes rather than on the speakers. Another experiment demonstrated that a combination of NPB-DAA and DSAE-PB outperformed the available methods in phoneme and word discovery tasks involving speech signals with Japanese vowel sequences from multiple speakers.

show abstract

“…First, we briefly introduce NPB-DAA (Taniguchi et al, 2016b). Secondly, we describe DSAE-PBHL after introducing DSAE (Ng, 2011a;Liu et al, 2015;Taniguchi et al, 2016c).…”

Section: Methodsmentioning

confidence: 99%

Unsupervised Phoneme and Word Discovery From Multiple Speakers Using Double Articulation Analyzer and Neural Network With Parametric Bias

2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…Note that, we excluded PCA-FP as a comparison method because of its poor ability to extract latent features with defects in previous experiments. To segment the time-series of driving behaviors, we employed a sticky hierarchical Dirichlet process hidden Markov model (sticky HDP-HMM) [ 33 ], which was used for segmenting driving behavior data in previous studies [ 8 , 21 , 34 , 35 ]. The RAW data and time-series of latent features extracted from D12 were used to train the sticky HDP-HMM.…”

Section: Application: Driving Behavior Segmentation With Defectsmentioning

confidence: 99%

“…Although the above two studies included feature extraction for time-series analysis, those features were extracted by supervised learning models for respective tasks. We used an unsupervised learning method for feature extraction in this study, because, if the time-series of latent features extracted via an unsupervised learning method are able to represent driving behavior, they can support varied tasks including tasks based on the unsupervised method, e.g., time-series segmentation [ 21 ] and data visualization [ 22 ].…”

Section: Introductionmentioning

confidence: 99%

Defect-Repairable Latent Feature Extraction of Driving Behavior via a Deep Sparse Autoencoder

Liu

Taniguchi

Takenaka

et al. 2018

Sensors

Self Cite

View full text Add to dashboard Cite

Data representing driving behavior, as measured by various sensors installed in a vehicle, are collected as multi-dimensional sensor time-series data. These data often include redundant information, e.g., both the speed of wheels and the engine speed represent the velocity of the vehicle. Redundant information can be expected to complicate the data analysis, e.g., more factors need to be analyzed; even varying the levels of redundancy can influence the results of the analysis. We assume that the measured multi-dimensional sensor time-series data of driving behavior are generated from low-dimensional data shared by the many types of one-dimensional data of which multi-dimensional time-series data are composed. Meanwhile, sensor time-series data may be defective because of sensor failure. Therefore, another important function is to reduce the negative effect of defective data when extracting low-dimensional time-series data. This study proposes a defect-repairable feature extraction method based on a deep sparse autoencoder (DSAE) to extract low-dimensional time-series data. In the experiments, we show that DSAE provides high-performance latent feature extraction for driving behavior, even for defective sensor time-series data. In addition, we show that the negative effect of defects on the driving behavior segmentation task could be reduced using the latent features extracted by DSAE.

show abstract

“…Studies of language acquisition also constitute a constructive approach to the human developmental process (Cangelosi and Schlesinger, 2015 ), the language grounding (Steels and Hild, 2012 ), and the symbol emergence (Taniguchi et al, 2016c ). One approach to studying language acquisition focuses on the estimation of phonemes and words from speech signals (Goldwater et al, 2009 ; Heymann et al, 2014 ; Taniguchi et al, 2016d ). However, these studies used only continuous speech signals without using co-occurrence based on other sensor information, e.g., visual, tactile, and proprioceptive information.…”

Section: Related Workmentioning

confidence: 99%

Cross-Situational Learning with Bayesian Generative Models for Multimodal Category and Word Learning in Robots

Taniguchi

Cangelosi

2017

Front. Neurorobot.

Self Cite

View full text Add to dashboard Cite

In this paper, we propose a Bayesian generative model that can form multiple categories based on each sensory-channel and can associate words with any of the four sensory-channels (action, position, object, and color). This paper focuses on cross-situational learning using the co-occurrence between words and information of sensory-channels in complex situations rather than conventional situations of cross-situational learning. We conducted a learning scenario using a simulator and a real humanoid iCub robot. In the scenario, a human tutor provided a sentence that describes an object of visual attention and an accompanying action to the robot. The scenario was set as follows: the number of words per sensory-channel was three or four, and the number of trials for learning was 20 and 40 for the simulator and 25 and 40 for the real robot. The experimental results showed that the proposed method was able to estimate the multiple categorizations and to learn the relationships between multiple sensory-channels and words accurately. In addition, we conducted an action generation task and an action description task based on word meanings learned in the cross-situational learning scenario. The experimental results showed that the robot could successfully use the word meanings learned by using the proposed method.

show abstract

Double articulation analyzer with deep sparse autoencoder for unsupervised word discovery from speech signals

Cited by 27 publications

References 17 publications

Unsupervised Phoneme and Word Discovery From Multiple Speakers Using Double Articulation Analyzer and Neural Network With Parametric Bias

Unsupervised Phoneme and Word Discovery From Multiple Speakers Using Double Articulation Analyzer and Neural Network With Parametric Bias

Defect-Repairable Latent Feature Extraction of Driving Behavior via a Deep Sparse Autoencoder

Cross-Situational Learning with Bayesian Generative Models for Multimodal Category and Word Learning in Robots

Contact Info

Product

Resources

About