• Speech Sciences
• /
• v.10 no.3
• /
• pp.263-277
• /
• 2003

One of major problems in speech recognition is performance degradation due to the mismatch between the training and test environments. Recently, Stereo-based Piecewise LInear Compensation for Environments (SPLICE), which is frame-based bias removal algorithm for cepstral enhancement using stereo training data and noisy speech model as a mixture of Gaussians, was proposed and showed good performance in noisy environments. In this paper, we propose several methods to improve the conventional SPLICE. First we apply Cepstral Mean Subtraction (CMS) as a preprocessor to SPLICE, instead of applying it as a postprocessor. Secondly, to compensate residual distortion after SPLICE processing, two-stage SPLICE is proposed. Thirdly we employ phonetic information for training SPLICE model. According to experiments on the Aurora 2 database, proposed method outperformed the conventional SPLICE and we achieved a 50% decrease in word error rate over the Aurora baseline system.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.5
• /
• pp.222-228
• /
• 2010

In this paper, we propose a new approach to noise estimation incorporating spectral deviation with soft decision scheme to enhance the intelligibility of the degraded speech signal in non-stationary noisy environments. Since the conventional noise estimation technique based on soft decision scheme estimates and updates the noise power spectrum using a fixed smoothing parameter which was assumed in stationary noisy environments, it is difficult to obtain the robust estimates of noise power spectrum in non-stationary noisy environments that spectral characteristics of noise signal such as restaurant constantly change. In this paper, once we first classify the stationary noise and non-stationary noise environments based on the analysis of spectral deviation of noise signal, we adaptively estimate and update the noise power spectrum according to the classified noise types. The performances of the proposed algorithm are evaluated by ITU-T P. 862 perceptual evaluation of speech quality (PESQ) under various ambient noise environments and show better performances compared with the conventional method.

• MALSORI
• /
• no.62
• /
• pp.97-112
• /
• 2007

The speech recognizer in general operates in noisy acoustical environments. Many research works have been done to cope with the acoustical variations. Among them, the multiple-HMM model approach seems to be quite effective compared with the conventional methods. In this paper, we consider a multiple-model approach combined with the model compensation method and investigate the necessary number of the HMM model sets through noisy speech recognition experiments. By using the data-driven Jacobian adaptation for the model compensation, the multiple-model approach with only a few model sets for each noise type could achieve comparable results with the re-training method.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.2
• /
• pp.137-142
• /
• 2020

Many studies based on I-vector have been conducted in a variety of environments, from text-dependent short-utterance to text-independent long-utterance. In this paper, we propose a speaker verification system employing a combination of I-vector with Probabilistic Linear Discriminant Analysis (PLDA) and speaker embedding of Long Short Term Memory (LSTM) with attention mechanism in far-field and noisy environments. The LSTM model's Equal Error Rate (EER) is 15.52 % and the Attention-LSTM model is 8.46 %, improving by 7.06 %. We show that the proposed method solves the problem of the existing extraction process which defines embedding as a heuristic. The EER of the I-vector/PLDA without combining is 6.18 % that shows the best performance. And combined with attention-LSTM based embedding is 2.57 % that is 3.61 % less than the baseline system, and which improves performance by 58.41 %.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.439-451
• /
• 2021

We propose a deep learning-based beamformer combined with spectral subtraction for continuous speech recognition operating in noisy environments. Conventional beamforming systems were mostly evaluated by using pre-segmented audio signals which were typically generated by mixing speech and noise continuously on a computer. However, since speech utterances are sparsely uttered along the time axis in real environments, conventional beamforming systems degrade in case when noise-only signals without speech are input. To alleviate this drawback, we combine online beamforming algorithm and spectral subtraction. We construct a Continuous Speech Enhancement (CSE) evaluation set to evaluate the online beamforming algorithm in noisy environments. The evaluation set is built by mixing sparsely-occurring speech utterances of the CHiME3 evaluation set and continuously-played CHiME3 background noise and background music of MUSDB. Using a Kaldi-based toolkit and Google web speech recognizer as a speech recognition back-end, we confirm that the proposed online beamforming algorithm with spectral subtraction shows better performance than the baseline online algorithm.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.1
• /
• pp.39-46
• /
• 2019

In this paper, combining features is proposed as a way to enhance the classification accuracy of sounds under noisy environments using the CNN (Convolutional Neural Network) structure. A robust log Mel-filter bank using Wiener filter and PNCCs (Power Normalized Cepstral Coefficients) are extracted to form a 2-dimensional feature that is used as input to the CNN structure. An ebird database is used to classify 43 types of bird species in their natural environment. To evaluate the performance of the combined features under noisy environments, the database is augmented with 3 types of noise under 4 different SNRs (Signal to Noise Ratios) (20 dB, 10 dB, 5 dB, 0 dB). The combined feature is compared to the log Mel-filter bank with and without incorporating the Wiener filter and the PNCCs. The combined feature is shown to outperform the other mentioned features under clean environments with a 1.34 % increase in overall average accuracy. Additionally, the accuracy under noisy environments at the 4 SNR levels is increased by 1.06 % and 0.65 % for shop and schoolyard noise backgrounds, respectively.

• ETRI Journal
• /
• v.30 no.1
• /
• pp.89-100
• /
• 2008

We consider the feature recombination technique in a multiband approach to speaker identification and verification. To overcome the ineffectiveness of conventional feature recombination in broadband noisy environments, we propose a new subband feature recombination which uses subband likelihoods and a subband reliable-feature selection technique with an adaptive noise model. In the decision step of speaker recognition, a few very low unreliable feature likelihood scores can cause a speaker recognition system to make an incorrect decision. To overcome this problem, reliable-feature selection adjusts the likelihood scores of an unreliable feature by comparison with those of an adaptive noise model, which is estimated by the maximum a posteriori adaptation technique using noise features directly obtained from noisy test speech. To evaluate the effectiveness of the proposed methods in noisy environments, we use the TIMIT database and the NTIMIT database, which is the corresponding telephone version of TIMIT database. The proposed subband feature recombination with subband reliable-feature selection achieves better performance than the conventional feature recombination system with reliable-feature selection.

• KIEAE Journal
• /
• v.3 no.2
• /
• pp.33-36
• /
• 2003

The influence of the visual information on the sound perception would be considerable. Furthermore, if the sound perception ranges in noisiness or annoyance beyond the loudness, it will depend much more on the shape of the visual information. This paper aims to estimate the influence of the several kinds of visual information on the perception of road traffic noise by means of the psycho-acoustic test method. The findings of present study on the influence of visual information on subjective noise perception are summarized as follows: Presenting visual images of mild and comfortable scenery reduced the noise perception reaction at the less noisy environments not exceeding 65 dB(A). At highly noisy environments exceeding 65 dB(A), however, the noise perception can be reduced by strong image of waterfall. Even eliminating the road traffic image may be helpful. Visual image of waterfall reduced the noise perception at all levels. It is inferred that the road traffic noise perception can be effectively ameliorated by presenting strong and real landscape images at any noisy environment.

• International Journal of Internet, Broadcasting and Communication
• /
• v.11 no.4
• /
• pp.26-30
• /
• 2019

This paper proposes a new voice activity detection (VAD) method which is based on SNR and non-intrusive speech intelligibility estimation. In the conventional SNR-based VAD methods, voice activity probability is obtained by estimating frame-wise SNR at each spectral component. However these methods lack performance in various noisy environments. We devise a hybrid VAD method that uses non-intrusive speech intelligibility estimation as well as SNR estimation, where the speech intelligibility score is estimated based on deep neural network. In order to train model parameters of deep neural network, we use MFCC vector and the intrusive speech intelligibility score, STOI (Short-Time Objective Intelligent Measure), as input and output, respectively. We developed speech presence measure to classify each noisy frame as voice or non-voice by calculating the weighted average of the estimated STOI value and the conventional SNR-based VAD value at each frame. Experimental results show that the proposed method has better performance than the conventional VAD method in various noisy environments, especially when the SNR is very low.

• Proceedings of the IEEK Conference
• /
• 2003.07d
• /
• pp.1251-1254
• /
• 2003

In this paper, Experience Sensitive Cumulative Neural Network (ESCNN) is introduced, which can cumulate the same or similar experiences. As the same or similar training patterns are cumulated in the network, the system recognize more important information in the training patterns. The functions of forgetting less important informations and attending more important informations resided in the training patterns are surveyed and implemented by simulations. The system behaves well under the noisy circumstances due to its forgetting and/or attending properties, even in 50 percents noisy environments. This paper also describes the creation of the generalized patterns for the input training patterns.