• ETRI Journal
• /
• v.18 no.1
• /
• pp.29-41
• /
• 1996

This paper describes a method to select a suitable feature for speech recognition using information theoretic measure. Conventional speech recognition systems heuristically choose a portion of frequency components, cepstrum, mel-cepstrum, energy, and their time differences of speech waveforms as their speech features. However, these systems never have good performance if the selected features are not suitable for speech recognition. Since the recognition rate is the only performance measure of speech recognition system, it is hard to judge how suitable the selected feature is. To solve this problem, it is essential to analyze the feature itself, and measure how good the feature itself is. Good speech features should contain all of the class-related information and as small amount of the class-irrelevant variation as possible. In this paper, we suggest a method to measure the class-related information and the amount of the class-irrelevant variation based on the Shannon's information theory. Using this method, we compare the mel-scaled FFT, cepstrum, mel-cepstrum, and wavelet features of the TIMIT speech data. The result shows that, among these features, the mel-scaled FFT is the best feature for speech recognition based on the proposed measure.

• ETRI Journal
• /
• v.36 no.3
• /
• pp.514-517
• /
• 2014

In this paper, alternative dynamic features for speech recognition are proposed. The goal of this work is to improve speech recognition accuracy by deriving the representation of distinctive dynamic characteristics from a speech spectrum. This work was inspired by two temporal dynamics of a speech signal. One is the highly non-stationary nature of speech, and the other is the inter-frame change of a speech spectrum. We adopt the use of a sub-frame spectrum analyzer to capture very rapid spectral changes within a speech analysis frame. In addition, we attempt to measure spectral fluctuations of a more complex manner as opposed to traditional dynamic features such as delta or double-delta. To evaluate the proposed features, speech recognition tests over smartphone environments were conducted. The experimental results show that the feature streams simply combined with the proposed features are effective for an improvement in the recognition accuracy of a hidden Markov model-based speech recognizer.

• ETRI Journal
• /
• v.34 no.4
• /
• pp.583-590
• /
• 2012

In this paper, we present a new approach for speaker diarization. First, we use the prosodic information calculated on the original speech to resynthesize the new speech data utilizing the spectrum modeling technique. The resynthesized data is modeled with sinusoids based on pitch, vibration amplitude, and phase bias. Then, we use the resynthesized speech data to extract cepstral features and integrate them with the cepstral features from original speech for speaker diarization. At last, we show how the two streams of cepstral features can be combined to improve the robustness of speaker diarization. Experiments carried out on the standardized datasets (the US National Institute of Standards and Technology Rich Transcription 04-S multiple distant microphone conditions) show a significant improvement in diarization error rate compared to the system based on only the feature stream from original speech.

• Phonetics and Speech Sciences
• /
• v.15 no.2
• /
• pp.53-59
• /
• 2023

Detection of children with autism spectrum disorder (ASD) based on speech has relied on predefined feature sets due to their ease of use and the capabilities of speech analysis. However, clinical impressions may not be adequately captured due to the broad range and the large number of features included. This paper demonstrates that the knowledge-driven speech features (KDSFs) specifically tailored to the speech traits of ASD are more effective and efficient for detecting speech of ASD children from that of children with typical development (TD) than a predefined feature set, extended Geneva Minimalistic Acoustic Standard Parameter Set (eGeMAPS). The KDSFs encompass various speech characteristics related to frequency, voice quality, speech rate, and spectral features, that have been identified as corresponding to certain of their distinctive attributes of them. The speech dataset used for the experiments consists of 63 ASD children and 9 TD children. To alleviate the imbalance in the number of training utterances, a data augmentation technique was applied to TD children's utterances. The support vector machine (SVM) classifier trained with the KDSFs achieved an accuracy of 91.25%, surpassing the 88.08% obtained using the predefined set. This result underscores the importance of incorporating domain knowledge in the development of speech technologies for individuals with disorders.

• Phonetics and Speech Sciences
• /
• v.3 no.4
• /
• pp.85-91
• /
• 2011

Sasang constitution medicine utilizes voice characteristics to diagnose a person's constitution. To classify Sasang constitutional groups using speech information technology, this study aims at establishing the relationship between Sasang constitutional groups and their corresponding voice characteristics by investigating various speech feature variables. The speech variables include features related to speech source and vocal tract filter. Experimental results show that statistically significant correlation between voice characteristics and some speech feature variables is observed.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.2E
• /
• pp.71-76
• /
• 2005

Compared to normal speech, distant-talking speech is characterized by the acoustic effect due to interfering sound and echoes as well as articulatory changes resulting from the speaker's effort to be more intelligible. In this paper, the acoustic features for distant-talking speech due to the articulatory changes will be analyzed and compared with those of the Lombard effect. In order to examine the effect of different distances and articulatory changes, speech recognition experiments were conducted for normal speech as well as distant-talking speech at different distances using HTK. The speech data used in this study consist of 4500 distant-talking utterances and 4500 normal utterances of 90 speakers (56 males and 34 females). Acoustic features selected for the analysis were duration, formants (F1 and F2), fundamental frequency, total energy and energy distribution. The results show that the acoustic-phonetic features for distant-talking speech correspond mostly to those of Lombard speech, in that the main resulting acoustic changes between normal and distant-talking speech are the increase in vowel duration, the shift in first and second formant, the increase in fundamental frequency, the increase in total energy and the shift in energy from low frequency band to middle or high bands.

• Phonetics and Speech Sciences
• /
• v.4 no.2
• /
• pp.73-78
• /
• 2012

Emotion recognition is an important technology in the filed of human-machine interface. To apply speech technology to emotion recognition, this study aims to establish a relationship between emotional groups and their corresponding voice characteristics by investigating various speech features. The speech features related to speech source and vocal tract filter are included. Experimental results show that statistically significant speech parameters for classifying the emotional groups are mainly related to speech sources such as jitter, shimmer, F0 (F0_min, F0_max, F0_mean, F0_std), harmonic parameters (H1, H2, HNR05, HNR15, HNR25, HNR35), and SPI.

• Phonetics and Speech Sciences
• /
• v.4 no.4
• /
• pp.71-77
• /
• 2012

Sasang constitution experts utilize voice characteristics as an auxiliary measure for deciding a person's constitutional group. This study aims at establishing a relationship between speech features and the constitutional groups by subjective listening evaluation of voice characteristics. A speech database of 841 speakers whose constitutional groups have been already diagnosed by Sasang constitution experts was constructed. Speech features related to speech source and vocal tract filter were extracted from five vowels and one sentence. Statistically significant speech features for classifying the groups were analyzed using SPSS. The features contributed to constitution classification were speaking rate, Energy, A1, A2, A3, H1, H2, H4, CPP for males in their 20s, F0_mean, CPP, SPI, HNR, Shimmer, Energy, A1, A2, A3, H1, H2, H4 for females in their 20s, Energy, A1, A2, A3, H1, H2, H4, CPP for male in the 60s, and Jitter, HNR, CPP, SPI for females in their 60s. Experimental results show that speech technology is useful in classifying constitutional groups.

• ETRI Journal
• /
• v.44 no.3
• /
• pp.462-475
• /
• 2022

Although researchers have proposed numerous techniques for speech emotion recognition, its performance remains unsatisfactory in many application scenarios. In this study, we propose a speech emotion recognition model based on a genetic algorithm (GA)-decision tree (DT) fusion of deep and acoustic features. To more comprehensively express speech emotional information, first, frame-level deep and acoustic features are extracted from a speech signal. Next, five kinds of statistic variables of these features are calculated to obtain utterance-level features. The Fisher feature selection criterion is employed to select high-performance features, removing redundant information. In the feature fusion stage, the GA is is used to adaptively search for the best feature fusion weight. Finally, using the fused feature, the proposed speech emotion recognition model based on a DT support vector machine model is realized. Experimental results on the Berlin speech emotion database and the Chinese emotion speech database indicate that the proposed model outperforms an average weight fusion method.

• Journal of Korea Multimedia Society
• /
• v.21 no.7
• /
• pp.737-744
• /
• 2018

This paper presents a study on the performance of the higher-order moments for speech/music classification. For a successful speech/music classifier, extracting features that allow direct access to the relevant speech or music specific information is crucial. In addition to the conventional variance-based features, we utilize the higher-order moments of features, such as skewness and kurtosis. Moreover, we investigate the subband decomposition parameters in extracting features, which improves classification accuracy. Experiments on two speech/music datasets, which are publicly available, were performed and show that the higher-order moment features can improve classification accuracy when combined with the conventional variance-based features.