• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.454-460
• /
• 2020

Artificial intelligence assistants that provide speech recognition operate through cloud-based voice recognition with high accuracy. In cloud-based speech recognition, Wake-Up-Word (WUW) detection plays an important role in activating devices on standby. In this paper, we compare the performance of Convolutional Neural Network (CNN)-based WUW detection models for mobile devices by using Google's speech commands dataset, using the spectrogram and mel-frequency cepstral coefficient features as inputs. The CNN models used in this paper are multi-layer perceptron, general convolutional neural network, VGG16, VGG19, ResNet50, ResNet101, ResNet152, MobileNet. We also propose network that reduces the model size to 1/25 while maintaining the performance of MobileNet is also proposed.

• Journal of Digital Convergence
• /
• v.11 no.6
• /
• pp.169-174
• /
• 2013

Speech recognition, the problem of performance degradation is the difference between the model training and recognition environments. Silence features normalized using the method as a way to reduce the inconsistency of such an environment. Silence features normalized way of existing in the low signal-to-noise ratio. Increase the energy level of the silence interval for voice and non-voice classification accuracy due to the falling. There is a problem in the recognition performance is degraded. This paper proposed a robust speech detection method in noisy environments using a silence feature normalization and voice energy maximize. In the high signal-to-noise ratio for the proposed method was used to maximize the characteristics receive less characterized the effects of noise by the voice energy. Cepstral feature distribution of voice / non-voice characteristics in the low signal-to-noise ratio and improves the recognition performance. Result of the recognition experiment, recognition performance improved compared to the conventional method.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1E
• /
• pp.82-88
• /
• 1994

It is important to exactly detect the GCI(Glottal Closure Instant) in the speech signal processing. A few methods to detect the GCI of voiced speech have een proposer, untill now. But these are difficult to detect the GCI for wide range of speakers and or various vowel signals. In this paper, we prposed a new method for GCI detection using the G-peak. The speech waveforms are passed through the LPF of variable bandwidth. Then, the GCI's of voiced speech are detected by the G-peak based on the filtered signals. We compared the detected with the eye-checked GCI at the SNR of clean, 20dB, and 0dB. We took into account the range within 1ms between eye-checked and detected GCI. We obtained the result of the detection rate as 97.9% in the clean speech, 96.5% in 20dB SNR, and 94.8% in 0dB SNR, respectively.

• Speech Sciences
• /
• v.15 no.2
• /
• pp.55-65
• /
• 2008

Recently, the multi-model based speech recognizer has been used quite successfully for noisy speech recognition. For the selection of the reference HMM (hidden Markov model) which best matches the noise type and SNR (signal to noise ratio) of the input testing speech, the estimation of the SNR value using the VAD (voice activity detection) algorithm and the classification of the noise type based on the GMM (Gaussian mixture model) have been done separately in the multi-model framework. As the SNR estimation process is vulnerable to errors, we propose an efficient method which can classify simultaneously the SNR values and noise types. The KL (Kullback-Leibler) distance between the single Gaussian distributions for the noise signal during the training and testing is utilized for the classification. The recognition experiments have been done on the Aurora 2 database showing the usefulness of the model compensation method in the multi-model based speech recognizer. We could also see that further performance improvement was achievable by combining the probability density function of the MCT (multi-condition training) with that of the reference HMM compensated by the D-JA (data-driven Jacobian adaptation) in the multi-model based speech recognizer.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.633-638
• /
• 2003

In this paper, we propose a humanoid active audition system which detects the direction of sound and performs speech recognition using just three microphones. Compared with previous researches, this system which has simpler algorithm, fewer microphones and better amplifier shows better performance. In order to verify our system's performance, we install the proposed active audition system to the home service robot, called Hombot II, which has been developed at the KIST (Korea Institute of Science and Technology), thus we confirm excellent performance by experimental results

• Proceedings of the KIEE Conference
• /
• 1995.11a
• /
• pp.149-153
• /
• 1995

In this paper, wavelet transform with multi-resolution property is used to improve the accuracy of pitch estimation of speech signal. Pitch detection of speech signal is based on the local maxima by using wavelet transform. The wavelet transform of a signal is a multiscale decomposition that is well localized in space and frequency. The proposed pitch defection algorithm is suitable for both low-pitched and high-pitched speakers.

• Journal of the Institute of Convergence Signal Processing
• /
• v.7 no.3
• /
• pp.116-121
• /
• 2006

In this paper, we propose the dimension reduction method of multi-dimension speech feature vector for real-time adaptation procedure in various noisy environments. This method which reduces dimensions non-linearly to map the likelihood of speech feature vector and noise feature vector. The LRT(Likelihood Ratio Test) is used for classifying speech and non-speech. The results of implementation are similar to multi-dimensional speech feature vector. The results of speech recognition implementation of detected speech data are also similar to multi-dimensional(10-order dimensional MFCC(Mel-Frequency Cepstral Coefficient)) speech feature vector.

• Speech Sciences
• /
• v.11 no.3
• /
• pp.157-164
• /
• 2004

Human has average pitch-level when speak naturally. That is 'Habitual pitch level'. However, if noise added at speech, the pitch-wave is changed irregularly. We can estimate noise level of speech by using this point. This paper calculates energy level of the input speech, pitch period from of above limited energy level by NAMDF (Normalized Average Magnitude Difference Function) method, after cut each frame by pitch period unit, and propose a method that estimate noise level through closed pitch of input speech.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.50 no.12
• /
• pp.590-596
• /
• 2001

The pre-emphasis is an essential process for speech signal processing. Widely used two methods are the typical method using a fixed value near unity and te optimal method using the autocorrelation ratio of the signal. This study proposes a new pre-emphasis method using the short-term energy difference of speech signal, which can effectively compensate the glottal source characteristics and lip radiation characteristics. Using the proposed pre-emphasis, speech analysis, such as spectrum estimation, formant detection, is performed and the results are compared with those of the conventional two pre-emphasis methods. The speech analysis with 5 single vowels showed that the proposed method enhanced the spectral shapes and gave nearly constant formant frequencies and could escape the overlapping of adjacent two formants. comparison with FFT spectra had verified the above results and showed the accuracy of the proposed method. The computational complexity of the proposed method reduced to about 50% of the optimal method.

• Proceedings of the KSPS conference
• /
• 2003.10a
• /
• pp.37-40
• /
• 2003

This paper presents the preliminary result of the automatic pronunciation scoring for non-native English speakers, and shows the developmental process for an English speech recognizer for the educational and evaluational purposes. The proposed speech recognizer, featuring two refined acoustic model sets, implements the noise-robust data compensation, phonetic alignment, highly reliable rejection, key-word and phrase detection, easy-to-use language modeling toolkit, etc., The developed speech recognizer achieves 0.725 as the average correlation between the human raters and the machine scores, based on the speech database YOUTH for training and K-SEC for test.