• Proceedings of the KSPS conference
• /
• 2003.10a
• /
• pp.101-104
• /
• 2003

In this paper, speech quality is improved by removing abrupt noise intervals and then substituting the gaps with estimates of the previous speech waveform. An abrupt noise detection signal has been proposed as a prediction error signal by utilizing LP coefficients of the previous frame. Abrupt noise intervals are estimated by using spectral energy. After removing estimated noise intervals, we applied several waveform substitution techniques such as zero substitution, previous frame repetition, pattern matching, and pitch waveform replication. To prove the validity of our algorithm, the LPC spectral distortion test and the recognition test are executed and, the results show that the speech quality is fairly well improved.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.08a
• /
• pp.87-92
• /
• 1998

In general, the synthesis unit database was constructed by recording isolated word. In that case, each boundary of word has typical prosodic pattern like a falling intonation or preboundary lengthening. To get natural synthetic speech using these kinds of database, we must artificially distort original speech. However, that artificial process rather resulted in unnatural, unintelligible synthetic speech due to the excessive prosodic modification on speech signal. To overcome these problems, we gathered thousands of sentences for synthesis database. To make a phone level synthesis unit, we trained speech recognizer with the recorded speech, and then segmented phone boundaries automatically. In addition, we used laryngo graph for the epoch detection. From the automatically generated synthesis database, we chose the best phone and directly concatenated it without any prosody processing. To select the best phone among multiple phone candidates, we used prosodic information such as break strength of word boundaries, phonetic contexts, cepstrum, pitch, energy, and phone duration. From the pilot test, we obtained some positive results.

• Speech Sciences
• /
• v.4 no.2
• /
• pp.55-68
• /
• 1998

In this paper, a novel method of noise reduction of speech based on a complex adaptive noise canceler and method of optimal feature extraction are proposed. This complex adaptive noise canceler needs simply the noise detection, and LMS algorithm used to calculate the adaptive filter coefficient. The method of optimal feature extraction requires the variance of noise. The experimental results have shown that the proposed method effectively reduced noise in noisy speech. Optimal feature extraction has shown similar characteristics in noise-free speech.

• Annual Conference on Human and Language Technology
• /
• 1999.10e
• /
• pp.335-342
• /
• 1999

The experiments reported here investigated whether seeing a speaker's face (visible speech) affects the perception and memory of Korean speech sounds. In order to exclude the possibility of top-down, knowledge-based influences on perception and memory, the experiments tested people with no knowledge of Korean. The first experiment examined whether visible speech (Auditory and Visual - AV) assists English native speakers (with no knowledge of Korean) in the detection of a syllable within a Korean speech phrase. It was found that a syllable was more likely to be detected within a phrase when the participants could see the speaker's face. The second experiment investigated whether English native speakers' judgments about the duration of a Korean phrase would be affected by visible speech. It was found that in the AV condition participant's estimates of phrase duration were highly correlated with the actual durations whereas those in the AO condition were not. The results are discussed with respect to the benefits of communication with multimodal information and future applications.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.4E
• /
• pp.161-166
• /
• 2003

We present an adaptive wavelet packet based speech enhancement method with robust voice activity detection (VAD) in non-stationary noise environment. The proposed method can be divided into two main procedures. The first procedure is a VAD with adaptive wavelet packet transform. And the other is a speech enhancement procedure based on the proposed VAD method. The proposed VAD method shows remarkable performance even in low SNRs and non-stationary noise environment. And subjective evaluation shows that the performance of the proposed speech enhancement method with wavelet bases is better than that with Fourier basis.

• Proceedings of the KOSOMBE Conference
• /
• v.1993 no.05
• /
• pp.90-93
• /
• 1993

In this study, Korean speech was recognized by using spatio-temporal recognition neural network. The subjects of speech are numeric speech from zero to nine and basic command which might be used for motorized wheelchair developed it own Lab. Rabiner and Sambur's method of speech detection was used in determining end-point of speech, speech parameter was extracted by using LPC 16 order. The recognition rate was over 90%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2009.10a
• /
• pp.705-708
• /
• 2009

This research describes a neural network-based phoneme segmenter for recognizing spontaneous speech. The input of the phoneme segmenter for spontaneous speech is 16th order mel-scaled FFT, normalized frame energy, ratio of energy among 0~3[KHz] band and more than 3[KHz] band. All the features are differences of two consecutive 10 [msec] frame. The main body of the segmenter is single-hidden layer MLP(Multi-Layer Perceptron) with 72 inputs, 20 hidden nodes, and one output node. The segmentation accuracy is 78% with 7.8% insertion.

• ETRI Journal
• /
• v.46 no.1
• /
• pp.96-105
• /
• 2024

We propose a method to automatically predict Alzheimer's disease from speech data using the ChatGPT large language model. Alzheimer's disease patients often exhibit distinctive characteristics when describing images, such as difficulties in recalling words, grammar errors, repetitive language, and incoherent narratives. For prediction, we initially employ a speech recognition system to transcribe participants' speech into text. We then gather opinions by inputting the transcribed text into ChatGPT as well as a prompt designed to solicit fluency evaluations. Subsequently, we extract embeddings from the speech, text, and opinions by the pretrained models. Finally, we use a classifier consisting of transformer blocks and linear layers to identify participants with this type of dementia. Experiments are conducted using the extensively used ADReSSo dataset. The results yield a maximum accuracy of 87.3% when speech, text, and opinions are used in conjunction. This finding suggests the potential of leveraging evaluation feedback from language models to address challenges in Alzheimer's disease recognition.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.2
• /
• pp.173-181
• /
• 2002

Generally the one of serious problems in Voice Activity Detection (VAD) is speech region detection in noise environment. Therefore, this paper propose the new method using energy, lsp varation. As a result of processing time and speech quality of the proposed algorithm, the processing time is reduced due to the accurate detection of inactive period, and there is almot no difference in the subjective quality test. As a result of bit rate, proposed algorithm measures the number of VAD=1 and the result shows predominant reduction of bit rate as SNR of noisy speech is low (about 5∼10 dB).

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1993.06a
• /
• pp.249-252
• /
• 1993

In speech signal processing, The accurate detection of the voiced/unvoiced is important for robust word recognition and analysis. This algorithm is based on the MD in the frame of speech signals that does not require statistical information about either signal or background-noise to decide a voiced/unvoiced. This paper presents a method of estimation the Characteristic of Magnitude Distribution from noisy speech and also of estimation the optimal threshold based on the MD of the voiced/unvoiced decision. The performances of this detectors is evaluated and compared to that obtained from classifying other paper.