We propose an MPEG-2 to H.264 transcoding method for interlace streams intermingled with frame and field macroblocks. This method uses the encoding information from an MPEG-2 stream and keeps as many DCT coefficients of the original MPEG-2 bitstream as possible. Experimental results show that the proposed method improves PSNR by about 0.190.31 dB compared with a conventional method.

To ensure a satisfactory QoE (Quality of Experience) and facilitate system design in speech recognition services, it is essential to establish a method that can be used to efficiently investigate recognition performance in different noise environments. Previously, we proposed a performance estimation method using a spectral distortion measure. However, there is the problem that recognition task complexity affects the relationship between the recognition performance and the distortion value. To solve this problem, this paper proposes a novel performance estimation method considering the recognition task complexity. We confirmed that the proposed method gives accurate estimates of the recognition performance for various recognition tasks by an experiment using noisy speech data recorded in a real room. Index Terms: performance estimation, noisy speech recognition, recognition task difficulty

Musical noise is a typical problem with blind source separation using a time-frequency mask. Recently, the cepstral smoothing of spectral masks (CSM) was proposed. Based on the idea of smoothing in the cepstral domain, this paper proposes the cepstral smoothing of separated signals (CSS) on the assumption that a cepstral representation better reflects the characteristics of speech signals than those of masks (or filter gains). We also report a comparative evaluation study of CSM and CSS with other musical noise reduction methods. Our experimental results show that CSM is effective for musical noise reduction, but the target speech was relatively distorted. On the other hand, our proposed CSS produced less distorted target signals with the same musical noise reduction as CSM.

A new method is described for quantifying the quality degradation introduced by wide-band speech codecs via a one-dimensional impairment factor. The method is based on auditory listening-only tests, but the resulting impairment factors may be used for predicting speech quality in an instrumental way, e.g., for network planning purposes. Following the method, auditory test results are first transformed to an overall quality rating scale, and then adjusted to rule out test-specific effects. The derived impairment factors fit into the common framework which is defined by the E-model for narrow-band telephone networks, and which is hereby extended towards wide-band speech transmission. This paper presents the necessary auditory test data, describes the derivation and adjustment methodology, and provides numerical values for a range of wide-band speech codecs. The values are tested for their robustness in case of codec tandems and adjusted to represent the effects of packet loss

It is essential to ensure quality of service (QoS) when offering a speech recognition service for use in noisy environments. This means that the recognition performance in the target noise environment must be investigated. One approach is to estimate the recognition performance from a distortion value, which represents the difference between noisy speech and its original clean version. Previously, estimation methods using the segmental signal-to-noise ratio (SNRseg), the cepstral distance (CD), and the perceptual evaluation of speech quality (PESQ) have been proposed. However, their estimation accuracy has not been verified for the case when a noise reduction algorithm is adopted as a preprocessing stage in speech recognition. We, therefore, evaluated the effectiveness of these distortion measures by experiments using the AURORA-2J connected digit recognition task and four different noise reduction algorithms. The results showed that in each case the distortion measure correlates well with the word accuracy when the estimators used are optimized for each individual noise reduction algorithm. In addition, it was confirmed that when a single estimator, optimized for all the noise reduction algorithms, is used, the PESQ method gives a more accurate estimate than SNRseg and CD. Furthermore, we have proposed the use of artificial voice of several seconds duration instead of a large amount of real speech and confirmed that a relatively accurate estimate can be obtained by using the artificial voice

During the last decade, a number of noise reduction algo-rithms were proposed for realizing noise robust speech recognition. However, their effectiveness strongly de-pends on noise conditions. One way for solving this problem is to select an optimal algorithm every time be-fore or after recognition process. This paper proposes a new method for integrating N-best recognition results ob-tained by multiple noise reduction algorithms. The pro-posed method selects the best recognition result by us-ing a confidence measure based on a frame-normalized log likelihood score. To evaluate the performance of the proposed method, recognition experiments were per-formed on the AURORA-2J connected digit recognition task. These results confirmed that the proposed method is very effective in the high and middle SNR conditions. 1.

A method of estimating the number of sound sources in a reverberant sound field is proposed in this paper. It is known that the eigenvalue distribution of the spatial correlation matrix calculated from a multiple microphone input reflects information on the number of sources. However, in a reverberant sound field, the feature of the number of sources in the eigenvalue distribution is degraded by the room reverberation. In this paper, support vector machines is applied to classify the eigenvalue distributions which are not clearly separable. The proposed method is then applied to the source separation system and is evaluated via automatic speech recognition.

Two array signal processing techniques are combined with independent component analysis (ICA) to enhance the performance of blind separation of acoustic signals in a reflective environment. The first technique is the subspace method which reduces the effect of room reflection when the system is used in a room. Room reflection is one of the biggest problems in blind source separation (BSS) in acoustic environments. The second technique is a method of solving permutation. For employing the subspace method, ICA must be used in the frequency domain, and precise permutation is necessary for all frequencies. In this method, a physical property of the mixing matrix, i.e., the coherency in adjacent frequencies, is utilized to solve the permutation. The experiments in a meeting room showed that the subspace method improved the rate of automatic speech recognition from 50% to 68% and that the method of solving permutation achieves performance that closely approaches that of the correct permutation, differing by only 4% in recognition rate.

Two array signal processing techniques are combined with independent component analysis to enhance the performance of blind separation of acoustic signals in a reflective environment such as rooms. The first technique is the subspace method which reduces the effect of room reflection. The second technique is a method of solving the permutation, in which the coherency of the mixing matrix in adjacent frequencies is utilized.

A method is proposed for modeling a room transfer function (RTF) by using common acoustical poles and their residues. The common acoustical poles correspond to the resonance frequencies (eigenfrequencies) of the room, so they are independent of the source and receiver positions. The residues correspond to the eigenfunctions of the room. Therefore, the residue, which is a function of the source and receiver positions, can be expressed using simple analytical functions for rooms with a simple geometry such as rectangular. That is, the proposed model can describe RTF variations using simple residue functions. Based on the proposed common-acoustical-pole and residue model, methods are also proposed for spatially interpolating and extrapolating RTFs. Because the common acoustical poles are invariant in a given room, the interpolation or extrapolation of RTFs is reformulated as a problem of interpolating or extrapolating residue values. The experimental results for a rectangular room, in which the residue values are interpolated or extrapolated by using a cosine function or a linear prediction method, demonstrate that unknown RTFs can be well estimated at low frequencies from known (measured) RTFs by using the proposed methods.

Use of a common-acoustical-pole and zero model is proposed for modeling head-related transfer functions (HRTFs) for various directions of sound incidence. The HRTFs are expressed using the common acoustical poles, which do not depend on the source directions, and the zeros, which do. The common acoustical poles are estimated as they are common to HRTFs for various source directions; the estimated values of the poles agree well with the resonance frequencies of the ear canal. Because this model uses only the zeros to express the HRTF variations due to changes in source direction, it requires fewer parameters (the order of the zeros) that depend on the source direction than do the conventional all-zero or pole/zero models. Furthermore, the proposed model can extract the zeros that are missed in the conventional models because of pole-zero cancellation. As a result, the directional dependence of the zeros can be traced well. Analysis of the zeros for HRTFs on the horizontal plane showed that the nonminimum-phase zero variation was well formulated using a simple pinna-reflection model. The common-acoustical-pole and zero (CAPZ) model is thus effective for modeling and analyzing HRTF's.

Describes a new artificial speech signal (ASVQ: artificial speech by vector quantization technique) which reflects the average characteristics of the human voice. The ASVQ is intended for use as a test signal in the objective evaluation of speech coding system quality. To obtain the average characteristics, a very large speech data base is analysed, The ASVQ generation method which reflects the extracted average characteristics of the human voice is formulated. This method applies vector quantizing analysis to the speech data base. The LPC speech synthesis circuit is used to reproduce the average characteristics. Finally, the new artificial speech signal is compared with a human voice and the estimation accuracy of the subjective quality of speech coding systems and nonlinear distortions is evaluated.>

An objective evaluation model called OPINE (overall performance index model for network evaluation) which is being studied for telephone transmission quality is described. The model estimates an MOS (mean opinion score) by summing psychological values on independent psychological factors, given the physical values of fundamental listening factors such as loss and noise. For the model to incorporate new talking factors such as talker echo and sidestone, evaluation characteristics are studied when both listening and talking factors are present in a telephone transmission system. The interaction of the listening and talking factors is discussed, and it is shown experimentally that these factors are independent of fundamental listening factors. From these results, an extended OPINE is proposed by establishing new psychological factors: talker echo and sidestone. The estimated MOS for composite conditions of listening and talking factors agrees well within the confidence interval of subjective MOS deviation.>

The effect of transmission delay on speech quality in telecommunications is described, with human factors such as conversational mode and the talker's knowledge of the cause of delay taken into account. Objective quality estimation methods for delay effects are proposed, and these methods are applied in an actual communications network. In connection with delay perception in a telephone conversation, the assumption was verified that a talker expects a particular response time from a partner, and that delay that is outside this expectation time window is noticed. Taking this information into account, a subjective conversational experiment is controlled by six kinds of tasks by varying the temporal characteristics. Thus, a subjective assessment of delay effects is obtained by laboratory tests in relation to the detectability threshold, opinion rating, and conversational efficiency. Objective quality measures for each test were defined as a linear combination of temporal parameters that correspond closely to subjective qualities.>

An LPC (linear predictive coding) cepstrum distance measure (CD) is introduced as an objective measure for estimating the subjective quality of speech signals. Good correspondence between LPC CD and the subjective quality, expressed in terms of both opinion equivalent Q and mean opinion score, are shown. Good repeatability of objective quality evaluation using LPC CD is also shown. A method for generating an artificial voice signal that reflects the characteristics of real speech signals is described. The LPC CD values calculated using this artificial voice are almost the same as those calculated using real speech signals. The speaker-dependency of the coded-speech quality is shown to be an important factor in low-bit-rate speech coding. Even taking this factor into consideration, LPC CD is shown to be effective for estimating the subjective quality.>

An adaptive predictive coding with adaptive bit allocation (APC-AB) is presented for speech encoding at low to medium bit rates (6.4kb/s-24kb/s). In this system, a split-band predictive coding scheme and a bit allocation scheme are employed in order to remove the redundancies due to a periodic concentration of the prediction residual energy as well as nonuniform nature of the speech spectrum. Quantization bits are dynamically allocated both over the sub-bands(frequency domain) and over the subintervals(time domain) in accordance with the distribution of the residual energies in the time-frequency domain. Optimum bit allocation is derived based on the mean square error criterion on the speech waveform, and the SNR gain is presented in relation to the prediction gain of the full-band signal. This system is evaluated in terms of the segmental SNR and speech quality. The result shows that the APC-AB system has advantage over the conventional full-band APC system in the segmental SNR and the stability of the prediction loop. It was also shown that this system can provide speech quality subjectively equivalent to 7 bit Log-PCM at 16 kb/s, and to 6 bit Log-PCM at 9.6 kb/s.

This paper describes objective quality measures to evaluate speech quality for various kinds of voiceband CODECs in common. The voiceband CODECs studied were PCM, ADM, ADPCM, ATC (Adaptive Transform Coding) and APC-AB (Adaptive Predictive Coding with Adaptive Bit Allocation). First, several objective quality measures in time and frequency domain were defined. They were SNR, Segmental SNR, Spectral Distortion, LPC Cepstrum Distance, COSH, Likelihood Ratio and Weighted Likelihood Ratio. Second, speech quality for voiceband CODECs were evaluated by subjective and objective quality measures. The subjective measures used were based on opinion test and articulation test. Finally, the relationship between objective measures and subjectively evaluated values was studied. It was concluded that the LPC Cepstrum Distance measure had best correspondence to Mean Opinion Score, among the objective measures studied. It was also concluded that the Wighted Likelihood Ratio measure had best correspondence to Articulation Score.