Conventional loss functions for gradient descent are designed mainly to assess output quality, with limited attention to gradient behavior. This study identifies the gradient inconsistency problem in existing loss functions, where gradient magnitude is misaligned with the loss value, causing unstable learning and poor correspondence between optimization and the target metric. To resolve this, we propose a gradient-aware loss function whose gradient is proportional to a chosen evaluation metric, with the loss derived through integration. This design ensures consistent gradients, stable convergence, and alignment with the evaluation metric. Applied to speech enhancement using the log-spectral distance metric, the method achieves faster convergence, greater stability, and higher perceptual quality than conventional approaches, validating the superiority of the proposed formulation.

This paper proposes a new method for pre-echo reduction in transform-based audio coding by controlling the temporal envelope of the waveform. The proposed method comprises two operating modes: temporal envelope flattening and temporal envelope correction of a target signal. The proposed method estimates signal levels with a low temporal resolution from side information using machine learning and converts them into a signal to be applied to the target signal to flatten and correct the temporal envelope. It also adjusts the signals to maintain signal continuity between the non-transient and transient frames. The proposed method differs from conventional methods in that it directly modifies the waveform before encoding and after decoding, which makes it useful as a new coding tool for legacy codecs. A subjective performance evaluation confirms that the proposed method uses fewer bits to provide sound quality equivalent to that of the short-window transform.1

This letter proposes a new method for audio coding that utilizes blind spectral recovery to improve the coding efficiency without compromising performance. The proposed method transmits only a fraction of the spectral coefficients, thereby reducing the coding bit rate. Then, it recovers the remaining coefficients in the decoder using the transmitted coefficients as input. The proposed method is differentiated from conventional spectral recovery in that the coefficients to be recovered are interleaved with the transmitted coefficients to obtain the most data correlation. Further, it enhances the transmitted coefficients, which are degraded by quantization errors, to deliver better information to the recovery process. The spectral recovery is conducted recursively on a band basis such that information recovered in one band is used for the recovery in subsequent bands. An improved level correction for the recovered coefficients and a new sign coding are also developed. A subjective performance evaluation confirms that the proposed method at 40 kbps provides statistically equivalent sound quality to a state-of-the-art coding method at 48 kbps for speech and music categories.

A process of spectral recovery can enhance the performance of transform-based audio coding by transmitting only a portion of spectral data and recovering the missing spectral data in the decoder. This study proposes an enhanced method of audio coding based on spectral recovery with an adaptive structure that yields improved sound quality compared with the previous method. The spectral data to be recovered are arranged in an adaptive pattern depending on the difficulty of recovery. In addition, according to the spectral characteristics, prior information associated with these spectral data is selectively transmitted that helps a neural network improve the performance of magnitude recovery. Prior information also provides the signs of recovered magnitudes. A subjective performance evaluation shows that, for mono coding without window switching at 40 kbps, the proposed coding method provides better sound quality than the conventional method on average.

This study proposes a new method of audio coding based on spectral recovery, which can enhance the performance of transform audio coding. An encoder represents spectral information of an input in a time-frequency domain and transmits only a portion of it so that the remaining spectral information can be recovered based on the transmitted information. A decoder recovers the magnitudes of missing spectral information using a convolutional neural network. The signs of missing spectral information are either transmitted or randomly assigned, according to their importance. By combining transmission and recovery of spectral information, the proposed method can enhance the coding performance, compared with conventional transform coding. The subjective performance evaluation shows that, for mono coding at 39.4 kbps, the proposed method provides higher sound quality than the USAC, by an average MUSHRA score of 8.5.

A new method of extracting acoustic features based on auditory spike code is proposed. An auditory spike code represents the acoustic activities created by the signal, similar to sound encoding of the human auditory system. In the proposed method, an auditory spike code of the signal is computed using a 64‐band Gammatone filterbank as the kernel functions. Then, for each spectral band, the sum and non‐zero counts of the auditory spike code are determined, and the features corresponding to the population and occurrence rate of the acoustic activities for each band are computed. In addition, the distribution of the acoustic activities on a time axis is analysed based on the histogram of time intervals between the adjacent acoustic activities, and the features for expressing temporal properties of the signal are extracted. The reconstruction accuracy of the auditory spike code is also measured as the features. Different from most conventional features obtained by complex statistical modelling or learning, the features by the proposed method can directly show specific acoustic characteristics contained in the signal. These features are applied to a music genre classification, and it is confirmed that they provide a performance comparable to state‐of‐the‐art features.

This letter proposes a new method of estimating the location of a predominant source in an amplitude-panned stereo signal with two sources. When the conventional method of location estimation is applied to an amplitude-panned multi-source sound, a serious estimation error occurs due to interference between sources. To solve this problem, the proposed method includes an embedded source separation based on non-negative matrix factorization, which first determines the initial estimate of source location, and then computes the basis matrix of the source using the initial location estimate. In this way, the proposed method can perform the source separation without a training stage. The comparative evaluation confirms that the proposed method provides higher performance in location estimation than the conventional method with a training stage.

In this paper, an adaptive compressed sensing is proposed in order to enhance the performance of fast tetrahertz reflection tomography. The proposed method first acquires data at random measurement points in the spatial domain, and estimates the regions in each tomographic image where much degradation is expected. Then, it allocates additional measurement points to those regions, so that more data are acquired adaptively at the regions prone to degradation, thereby improving the quality of the reconstructed tomographic images. The proposed method was applied to the T-ray reflection tomography system, and the image quality enhancement by the proposed method, compared to the conventional method, was verified for the same number of measurement points.

This letter proposes a new method of improving the speech quality of VoIP by assigning a priority to each packet at the edge router of a DiffServ-based network. The proposed method determines the importance of a packet in the parameter domain by estimating the quality degradation caused by its loss and assigns a priority to the packet according to its importance, which ensures that important packets are more robust during transmission. The subjective and objective evaluation shows that the proposed method improves the VoIP speech quality in an environment prone to packet loss.

This letter proposes a new method of improving perceptual quality of speech in a noisy environment. Based on the perception theory that the temporal envelope plays a major role in assessing intelligibility, the proposed method uses a novel operation that enhances the fluctuation of the band-wise temporal envelope. It also contains pitch enhancement in frequency domain to further improve the perceptual quality. A new subjective evaluation scheme employing binaural listening is proposed in order to measure perceptual quality more reliably. The subjective evaluation shows that the proposed method improves perceptual quality of speech in various noisy environments.

An efficient enhanced variable rate codec (EVRC) codebook search method based on a two-stage search is proposed. At the first stage, a coarse codevector is selected by a fast sequential search, and at the second stage, the pulse replacement procedure is run to enhance the performance of selected codevector. Simulations with various speech data show that the proposed method yields voice quality equivalent to that by the standard method with only 23% codebook search load.

This paper describes "BI-CELP: baseline and implied CELP", which is a high quality speech coding method based on a code excited linear prediction (CELP) model employing excitation vectors combined from two codebooks, one from the baseline codebook and the other from the implied codebook. In this method the index of the baseline codebook is coded and transmitted to the receiver while the index of the implied codebook is extracted from the synthesized speech. This method has been applied to a lower rate voice coder at 8 kbit/s to produce high quality voice comparable to that of the 16 kbit/s G.728 LD-CELP. The performance of the 8 kbit/s BI-CELP coder is measured in terms of SNRseg and MOS. The average SNRseg is 12.14 dB which is 0.6 dB higher than that of the 8 kbit/s G.729 CS-ACELP. The MOS for the quiet input is 3.8 which is 0.02 higher than that of G.729 CS-ACELP. BI-CELP algorithm is implemented in real-time on a single TMS320C31 with 27 MIPS of CPU.

For image processing systems that have a limited size of region of support, say 3/spl times/3, direct implementation of morphological operations by a structuring element larger than the prefixed size is impossible. The decomposition of morphological operations by a large structuring element into a sequence of recursive operations, each using a smaller structuring element, enables the implementation of large morphological operations. In this paper, the authors present the decomposition of arbitrarily shaped (convex or concave) structuring elements into 3/spl times/3 elements, optimized with respect to the number of 3/spl times/3 elements. The decomposition is based on the concept of factorization of a structuring element into its prime factors. For a given structuring element, all its corresponding 3/spl times/3 prime concave factors are first determined. From the set of the prime factors, the decomposability of the structuring element is then established, and subsequently the structuring element is decomposed into a smallest possible set of 3/spl times/3 elements. Examples of optimal decomposition and structuring elements that are not decomposable are presented.>

A morphological operation using a large structuring element can be decomposed equivalently into a sequence of recursive operations, each using a smaller structuring element. However, an optimal decomposition of arbitrarily shaped structuring elements is yet to be found. In this paper, we have derived an optimal decomposition of a specific class of structuring elements/spl mdash/convex sets/spl mdash/for a specific type of machine/spl mdash/4-connected parallel array processors. The cost of morphological operation on 4-connected parallel array processors is the total number of 4-connected shifts required by the set of structuring elements. First, the original structuring element is decomposed into a set of prime factors, and then their locations are determined while minimizing the cost function. Proofs are presented to show the optimality of the decomposition. Examples of optimal decomposition are given and compared to an existing decomposition reported by Xu (1991).>