To address the challenges of low efficiency, centralization, and weak security in existing Internet of Drone authentication, this article proposes BASL-IoD, a blockchain-assisted lightweight scheme. The proposed system consists of a trusted authority, end users, unmanned aerial vehicles, and a blockchain-based distributed ground control station network. Smart contract is employed to enable dynamic access control. The authentication protocol is built upon bitwise XOR and one-way hash operations, while incorporating fuzzy extractor and physical unclonable function to enhance entity security. The scheme is formally proven secure under the Real-or-Random model, ensuring the semantic security of the session key, and is further validated using the AVISPA tool to confirms its resistance to MITM, replay, and impersonation attacks. Additional informal analysis demonstrates its resilience against various common attacks. BASL-IoD was experimentally tested in a wireless network environment, and its smart contract functionality was further verified through deployment in the Truffle and Ganache framework. Comparative evaluations with other competing schemes show that BASL-IoD provides stronger security, more comprehensive functionality, and significantly reduced computational, communication, and storage overhead.

Understanding multimodal humor and sarcasm detection remains a key challenge in artificial intelligence. Despite recent advances, inconsistencies in feature extraction, evaluation methods, and experimental setups have hindered fair comparisons across different approaches. To address this issue, we propose the Multimodal Humor and Sarcasm Detection Benchmark (MHSDB), the first unified evaluation platform specifically designed for these tasks. MHSDB combines four datasets in English and Hindi and standardizes feature extraction and evaluation processes to facilitate consistent comparisons. We systematically evaluate mainstream foundation models across audio, video, and text modalities. Unimodal representations are assessed using self-attention mechanisms, while multimodal representations are evaluated through mainstream fusion strategies, including utterance-level and sequence-level approaches. Our experimental results reveal that multimodal approaches outperform unimodal ones in capturing complex contexts and multi-layered semantics. Additionally, specific fusion strategies excel at integrating cross-modal information, achieving state-of-the-art performance, and paving the way for future research on optimizing feature representation and multimodal fusion.

Computational paralinguistics (ComParal) aims to develop algorithms and models to automatically detect, analyze, and interpret non-verbal information from speech communication, e. g., emotion, health state, age, and gender. Despite its rapid progress, it heavily depends on sophisticatedly designed models given specific paralinguistic tasks. Thus, the heterogeneity and diversity of ComParal models largely prevent the realistic implementation of ComParal models. Recently, with the advent of acoustic foundation models because of self-supervised learning, developing more generic models that can efficiently perceive a plethora of paralinguistic information has become an active topic in speech processing. However, it lacks a unified evaluation framework for a fair and consistent performance comparison. To bridge this gap, we conduct a large-scale benchmark, namelyParaLBench, which concentrates on standardizing the evaluation process of diverse paralinguistic tasks, including critical aspects of affective computing such as emotion recognition and emotion dimensions prediction, over different acoustic foundation models. This benchmark contains ten datasets with thirteen distinct paralinguistic tasks, covering short-, medium- and long-term characteristics. Each task is carried out on 14 acoustic foundation models under a unified evaluation framework, which allows for an unbiased methodological comparison and offers a grounded reference for the ComParal community. Based on the insights gained from ParaLBench, we also point out potential research directions, i. e., the cross-corpus generalizability, to propel ComParal research in the future. The code associated with this study will be available to foster the transparency and replicability of this work for succeeding researchers.

Directed acyclic graph (DAG) blockchain is a new paradigm of blockchain that solves the poor scalability and low throughput caused by a single-chain structure for IoT applications in the blockchain. However, for DAG-structured blockchains, it is difficult to determine the order of blocks. Additionally, there is also lack of secure and efficient ways to generate blocks in parallel. To address these problems, we propose an efficient DAG blockchain architecture. First, a novel heaviest chain rule based on the block weight is proposed to guide the selection of the parent and uncle blocks for a new block, which realize that appending is ordering. Second, a tree-based gossip protocol (TBGP) is proposed to improve consensus efficiency by reducing message redundancy. Furthermore, federated learning is used to select nodes constructing the tree-based gossip network (TBGN). Finally, compared to the random gossip protocol (RGP), the simulation results indicate TBGP can effectively reduce communication redundancy and improve consensus efficiency.

Speech emotion recognition is a vital contributor to the next generation of human-computer interaction (HCI). However, current existing small-scale databases have limited the development of related research. In this paper, we present LSSED, a challenging large-scale english speech emotion dataset, which has data collected from 820 subjects to simulate real- world distribution. In addition, we release some pre-trained models based on LSSED, which can not only promote the development of speech emotion recognition, but can also be transferred to related downstream tasks such as mental health analysis where data is extremely difficult to collect. Finally, our experiments show the necessity of large-scale datasets and the effectiveness of pre-trained models. The dateset will be released on https://github.com/tobefans/LSSED.

Affective social multimedia computing is an emergent research topic for both affective computing and multimedia research communities. Social multimedia is fundamentally changing how we communicate, interact, and collaborate with other people in our daily lives. Comparing with well-organized broadcast news and professionally made videos such as commercials, TV shows, and movies, social multimedia media computing imposes great challenges to research communities. Social multimedia contains much affective information. Effective extraction of affective information from social multimedia can greatly help social multimedia computing (e.g., processing, index, retrieval, and understanding). Although much progress have been made in traditional multimedia research on multimedia content analysis, indexing, and retrieval based on subjective concepts such as emotion, aesthetics, and preference, affective social multimedia computing is a new research area. The affective social multimedia computing aims to proceed affective information from social multi-media. For massive and heterogeneous social media data, the research requires multidisciplinary understanding of content and perceptual cues from social multimedia. This workshop served as a successful step towards this goal and attracted contributions from different research disciplines on the analysis of affective signals in interaction (multimodal analyses enabling artificial agents in Human-Machine Interaction, social Interaction with artificial agents) and social multimedia (e.g., twitter, wechat, weibo, youtube, facebook, etc). This paper provides a summary of the activities of the workshop and the accepted papers and abstracts.

The IEEE Spoken Language Technology Workshop (SLT) 2021 Alpha-mini Speech Challenge (ASC) is intended to improve research on keyword spotting (KWS) and sound source location (SSL) on humanoid robots. Many publications report significant improvements in deep learning based KWS and SSL on open source datasets in recent years. For deep learning model training, it is necessary to expand the data coverage to improve the model robustness. Thus, simulating multi-channel noisy and reverberant data from single-channel speech, noise, echo and room impulsive response (RIR) is widely adopted. However, this approach may generate mismatch between simulated data and recorded data in real application scenarios, especially echo data. In this challenge, we open source a sizable speech, keyword, echo and noise corpus for promoting data-driven methods, particularly deep-learning approaches on KWS and SSL. We also choose Alpha-mini, a humanoid robot produced by UBTECH equipped with a built-in four-microphone array on its head, to record development and evaluation sets under the actual Alpha-mini robot application scenario, including environ-mental noise as well as echo and mechanical noise generated by the robot itself for model evaluation. Furthermore, we illustrate the rules, evaluation methods and baselines for re-searchers to quickly assess their achievements and optimize their models.

Consensus is one of the key problems in blockchains. There are many articles analyzing the performance of threat models for blockchains. But the network stability seems lack of attention, which in fact affects the blockchain performance. This paper studies the performance of a well adopted consensus algorithm, Raft, in networks with non-negligible packet loss rate. In particular, we propose a simple but accurate analytical model to analyze the distributed network split probability. At a given time, we explicitly present the network split probability as a function of the network size, the packet loss rate, and the election timeout period. To validate our analysis, we implement a Raft simulator and the simulation results coincide with the analytical results. With the proposed model, one can predict the network split time and probability in theory and optimize the parameters in Raft consensus algorithm.

In this study, we focus on automated approaches to detect depression from clinical interviews using machine learning approached, which the models are trained on multi-modal data. Differentiating from successful machine learning approaches such as context-aware analysis through feature engineering and end-to-end deep neural networks to depression detection utilizing the Distress Analysis Interview Corpus, we propose a novel method that incorporates a data augmentation procedure based on topic modelling using transformer and deep 1D convolutional neural network (CNN) for acoustic feature modeling. The simulation results demonstrate the effectiveness of the proposed method for training multi-modal deep learning models. Our deep 1D CNN and transformer models achieve the state-of-the-art performance for the audio and text modalities respectively, while our multi-modal results are comparable with the state-of-the-art depression detection systems.

Affective social multimedia computing is an emergent research topic for both affective computing and multimedia research communities. Social multimedia is fundamentally changing how we communicate, interact, and collaborate with other people in our daily lives. Social multimedia contains much affective information. Effective extraction of affective information from social multimedia can greatly help social multimedia computing (e.g., processing, index, retrieval, and understanding). Besides, with the rapid development of digital photography and social networks, people get used to sharing their lives and expressing their opinions online. As a result, user-generated social media data, including text, images, audios, and videos, grow rapidly, which urgently demands advanced techniques on the management, retrieval, and understanding of these data.

In this paper, we aim at improving the performance of synthesized speech in statistical parametric speech synthesis (SPSS) based on a generative adversarial network (GAN). In particular, we propose a novel architecture combining the traditional acoustic loss function and the GAN's discriminative loss under a multi-task learning (MTL) framework. The mean squared error (MSE) is usually used to estimate the parameters of deep neural networks, which only considers the numerical difference between the raw audio and the synthesized one. To mitigate this problem, we introduce the GAN as a second task to determine if the input is a natural speech with specific conditions. In this MTL framework, the MSE optimization improves the stability of GAN, and at the same time GAN produces samples with a distribution closer to natural speech. Listening tests show that the multi-task architecture can generate more natural speech that satisfies human perception than the conventional methods.

This paper presents the techniques used in our contribution to Emotion Recognition in the Wild 2017 video based sub-challenge. The purpose of the sub-challenge is to classify the six basic emotions (angry, sad, happy, surprise, fear and disgust) and neutral. Our proposed solution utilizes three state-of-the-arts techniques to overcome the challenges for the wild emotion recognition. Deep network transfer learning is used for feature extraction. Spatial-temporal model fusion is to make full use of the complementary of different networks. Semi-auto reinforcement learning is for the optimization of fusion strategy based on dynamic outside feedbacks given by challenge organizers. The overall accuracy of the proposed approach on the challenge test dataset is 57.2%, which is better than the challenge baseline of 40.47% .

Automatic detection of pathological voice is a challenging task in speech processing. Appropriate acoustic cues of voice can be used to differentiate between normal voices and pathological voices. We propose a method to represent each speech utterance using three types of speech signal representations (i.e., cross-correlation matrix, Gaussian distribution and linear subspace) respectively. Various kernels were applied to these representations for measuring resemblance and difference. Four classifiers, i.e., KNN, kernel partial least squares, kernel SVM, and logistic regression, are studied for comparing their performance of classification. Finally, a simple fusion of learning classifiers from different acoustic representations was carried out at the score decision level for enhancing the performance. The different classifiers were evaluated on the Interspeech 2012 challenge development data set and test data set. Their effects in a fusion scheme are studied. The accuracy of the fusion system attained 78.0 % on test set, with an improved gain of 9.1 % over the challenge baseline 68.9 %.

Voice conversion (VC) aims to make one speaker (source) to sound like spoken by another speaker (target) without changing the language content. Most of the state-of-the-art voice conversion systems focus only on timbre conversion. However, the speaker identity is characterized by the source-related cues such as fundamental frequency and energy as well. In this work, we propose an exemplarbased sparse representation of timbre and prosody for voice conversion that does not necessitate separately timbre conversion and prosody conversions. The experiment results show that, in addition to the conversion of spectral features, the proper conversion of prosody features will improve the quality and speaker identity of the converted speech.

This paper presents the techniques used in our contribution to Emotion Recognition in the Wild 2016’s video based sub-challenge. The purpose of the sub-challenge is to classify the six basic emotions (angry, sad, happy, surprise, fear & disgust) and neutral. Compared to earlier years’ movie based datasets, this year’s test dataset introduced reality TV videos containing more spontaneous emotion. Our proposed solution is the fusion of facial expression recognition and audio emotion recognition subsystems at score level. For facial emotion recognition, starting from a network pre-trained on ImageNet training data, a deep Convolutional Neural Network is fine-tuned on FER2013 training data for feature extraction. The classifiers, i.e., kernel SVM, logistic regression and partial least squares are studied for comparison. An optimal fusion of classifiers learned from different kernels is carried out at the score level to improve system performance. For audio emotion recognition, a deep Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) is trained directly using the challenge dataset. Experimental results show that both subsystems individually and as a whole can achieve state-of-the art performance. The overall accuracy of the proposed approach on the challenge test dataset is 53.9%, which is better than the challenge baseline of 40.47% .

This paper is to show a representation of fundamental frequency (F0) using continuous wavelet transform (CWT) for prosody modeling in emotion conversion. Emotional conversion aims at converting speech from one emotion state to another. Specifically, we use CWT to decompose F0 into a five-scale representation that corresponds to five temporal scales. A neutral voice is converted to an emotional voice under an exemplar-based voice conversion framework, where both spectrum and F0 are simultaneously converted. The simulation results demonstrate that the dynamics of F0 in different temporal scales can be well captured and converted using the five-scale CWT representation. The converted speech signals are evaluated both objectively and subjectively, that confirm the effectiveness of the proposed method.

In this paper, we address the problem of phase retrieval to recover a signal from the magnitude of its Fourier transform. In many applications of phase retrieval, the signals encountered are naturally sparse. In this work, we consider the case where the signal is sparse under the assumption that few components are nonzero. We exploit further the sparse nature of the signals and propose a two stage sparse phase retrieval algorithm. A simple iterative minimization algorithm recovers a sparse signal from measurements of its Fourier transform (or other linear transform) magnitude based on the minimization of a block l1 norm. We show in the experiments that the proposed algorithm achieves a competitive performance. It is robust to noise and scalable in practical implementation. The proposed method converges to a more accurate and stable solution than other existing techniques for synthetic signals. For speech signals, experiments show that the voice quality of reconstructed speech signals is almost as good as the original signals.

Non-negative matrix factorization (NMF) aims at finding nonnegative representations of nonnegative data. Among different NMF algorithms, alternating direction method of multipliers (ADMM) is a popular one with superior performance. However, we find that ADMM shows instability and inferior performance on real-world data like speech signals. In this paper, to solve this problem, we develop a class of advanced regularized ADMM algorithms for NMF. Efficient and robust learning rules are achieved by incorporating l1-norm and the Frobenius norm regularization. The prior information of Laplacian distribution of data is used to solve the problem with a unique solution. We evaluate this class of ADMM algorithms using both synthetic and real speech signals for a source separation task at different cost functions, i.e., Euclidean distance (EUD), KullbackLeibler (KL) divergence and Itakura-Saito (IS) divergence. Results demonstrate that the proposed algorithms converge faster and yield more stable and accurate results than the original ADMM algorithm.

Pathological speech usually refers to the voice disorders resulting from atypicalities in voice and/or in the articulatory mechanisms due to disease, illness or other physical problem in the speech production system. It may increase unhealthy social behavior and voice abuse, and dramatically affect the patients' quality of life. Therefore, automatic intelligibility detection of pathological speech has an important role in the opportune treatment of pathological voices. This paper proposes to use asymmetric sparse kernel partial least squares classifier (ASKPLSC) for intelligibility detection of pathological speech. The proposed approach achieves an unweighted accuracy (UA) of 74.0%, which is 7.34% relative improvement of baseline system of an UA of 68.90% for the Pathology Sub-Challenge of INTERSPEECH 2012 Speaker Trait Challenge.

This paper focuses on the optimization of energy efficiency (EE) for BSs with distributed multiple-input multiple-output (MIMO) using an efficient antenna sleep strategy. A novel metric called sleep EE gain is proposed in this paper to evaluate the energy efficiency performance of sleep strategy in distributed MIMO systems. A related closed-form expression is developed to determine the optimum antenna sleep period and to optimize the proposed metric. Based on our analysis, we propose an efficient sleep strategy for distributed MIMO systems. Simulation results show that the proposed strategy can significantly improve the overall sleep energy efficiency gain of distributed MIMO systems.

This paper focuses on the optimization of coordinated spectrum sensing for cognitive radio (CR) networks using an efficient cognitive terminal (CT) assignment strategy. A metric called ineffective transmission-effective transmission ratio (ITETR) is proposed to evaluate the performance of the CT assignment strategy. Considering the diversity of channel utilization and CT sensing performance, the optimization of the proposed metric is formulated as an assignment problem. We provide a mathematical analysis as well as the related closed-form solution in the simplified case when all CTs have the same sensing performance for the same channel. Based on this analysis, an improved greedy algorithm-based strategy is proposed to determine the sub-optimal solution for the more general case, i.e., each CT exhibits different sensing performance for the same channel. Simulation results show that the proposed strategy can significantly improve the ITETR of CR networks when different channels possess different utilizations. When all channels have the same utilization, the proposed strategy yields a perfect approximation to the iterative Hungarian algorithm-based strategy while offering lower computation complexity.

In this paper, we compare the performance of support vector machine (SVM) and asymmetric SIMPLS classifiers in emotion recognition from naturalistic dialogues. These two classifiers are evaluated on the SEMAINE corpus that involves emotional binary classification tasks of four dimensions, namely, activation, expectation, power, and valence. The experimental results reveal that the asymmetric SIMPLS (ASIMPLS) classifier is less sensitive to class distribution, faster training processing, and higher classification accuracy on the average unweight recall for develop sets than the baseline. Using the develop set, we provide analysis and simulation-based insights about the selection of the number of components for model validation of the ASIMPLS classifier. For the test sets, the performance of the ASIMPLS classifier achieves an absolute improvement of 6.10%, 6.14%, 24.45%, 1.32% on the weighted recall value on above-mentioned four dimensions, respectively, over the baseline model.

In this paper, a novel MAC-layer spectrum sensing scheme base on continuous-time semi-Markov process is investigated for the purpose of improving spectrum sensing efficiency of cognitive radio (CR) systems. The scheme focuses on identification of the optimal sensing sequence of channel by modeling a group of licensed channels' usage pattern as a continuous-time semi-Markov process. Experimental results show that the proposed algorithm has the potential to achieve noticeably improved performance in terms of reduction the sensing overhead and second user's average throughput when compared to the conventional non-prioritization spectrum sensing approach, thus is suitable for CR networks.

Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

In this paper, we use a stochastic fixed-point theorem to study the stochastic convergence properties (in mean-squares sense) of the cascaded LMS predictor including conditions on the stepsize for the adaptive algorithm convergence and the misadjustment. An analytic expression for the misadjustment is derived for Gaussian statistical signals and shown to be exponentially dependent on the number of stages in the cascade structure, which is higher than the misadjustment of the conventional LMS filter. It can be observed that a higher misadjusment can be expected if the input signal is extremely uncorrelated.

This paper describes a method to increase speech intelligibility when the speech signal is being transmitted over telephone lines. In order to detect all factors which affect speech intelligibility, we use telephone simulation tool in ITUT Software Tools Library release 2005 (STL2005) to identify the most problematic telephone-channel deteriorations. Of the various effects considered, additive noise and bandwidth of telephone channel seem to have the greatest influence on human's recognition accuracy. To reduce this degradation, a Lombard production and perception effect model, which represents the spectral changes of speech signal uttered and perceived in noisy environment, is proposed. This model is realized by a threestage modification procedure involving time-scale, pitch-scale, the formants and the consonant-vowel energy ratio modifications of the normal speech signals. From listening test conducted, it is shown that the proposed compensation algorithm increases the intelligibility of speech in telephone channel environments at 0 dB Signal-to-Noise ratio (SNR) and -10 dB SNR by nearly 5.5% and 13% over the unprocessed signals respectively.

In this paper, we use a stochastic fixed-point theorem to analyze the stochastic convergence properties of the cascaded RLS-LMS prediction filter in terms of conditions of convergence and the misadjustment. It is shown that the cascaded RLS-LMS prediction filter converges to almost the same optimal solution of the conventional RLS filter. The misadjustment is shown to be exponentially dependent on the number of stages in the cascade structure and is higher than the misadjustment of the conventional RLS filter. However, the cascaded RLS-LMS prediction filter allows us to build up a low complexity RLS-like predictor with time-varying learning rate, which may be useful in uncertain and non-stationary environments.

This paper describes the cascaded recursive least square-least mean square (RLS-LMS) prediction, which is part of the recently published MPEG-4 Audio Lossless Coding international standard. The predictor consists of cascaded stages of simple linear predictors, with the prediction error at the output of one stage passed to the next stage as the input signal. A linear combiner adds up the intermediate estimates at the output of each prediction stage to give a final estimate of the RLS-LMS predictor. In the RLS-LMS predictor, the first prediction stage is a simple first-order predictor with a fixed coefficient value 1. The second prediction stage uses the recursive least square algorithm to adaptively update the predictor coefficients. The subsequent prediction stages use the normalized least mean square algorithm to update the predictor coefficients. The coefficients of the linear combiner are then updated using the sign-sign least mean square algorithm. For stereo audio signals, the RLS-LMS predictor uses both intrachannel prediction and interchannel prediction, which results in a 3% improvement in compression ratio over using only the intrachannel prediction. Through extensive tests, the MPEG-4 Audio Lossless coder using the RLS-LMS predictor has demonstrated a compression ratio that is on par with the best lossless audio coders in the field. In this paper, the structure of the RLS-LMS predictor is described in detail, and the optimal predictor configuration is studied through various experiments.

An FIR cascade structure for adaptive linear prediction is studied in which each stage FIR filter is independently adapted using an LMS algorithm. Theoretical analysis shows that the cascade performs a linear prediction in a way of successive refinement and each stage tries to obliterate the dominant mode of its input. Experimental results show that the performance of the cascade LMS predictor are in good agreement with our theoretical analysis.

In this paper, software tools which simulate the error correcting capability of forward error correction (FEC) codes, for the transmission of MPEG advanced audio coding (AAC) bitstreams over a wireless channel, are described. The tools consist of two applications. The first application is a channel transmission simulator with the features to set the choice of FEC codes, wireless channel characteristics and observe the simulation results. The second application is a graph plotting software that facilitates analysis of simulation results. Such a tool makes it convenient to investigate the transmission of audio data over wireless links.

We analyze the effect of the input signal correlation on the performance of an RLS-LMS based adaptive linear prediction algorithm for lossless audio compression. We show that the misadjustment is the same for both correlated and uncorrelated input samples. However, the variance of the residual error power of the algorithm increases as the signal variance increases, which may cause numerical instability or a stalling problem for the RLS algorithm. For real audio signals quantized at 16 bit and 24 bit, experimental results show that the algorithm is capable of modelling audio signals quantized at 16 bit, but it can yield reduced performance coding gain for audio signals quantized at 24 bit, which confirms our theoretical analysis.

We analyze the effect of both correlated and uncorrelated input signals on the performance of a cascade RLS-LMS based adaptive linear prediction algorithm for lossless compression of different resolution audio signals. We show that the sensitivity of the cascade RLS-LMS predictor to the input signal depends mainly on that of the RLS pre-whitener. Theoretical analysis shows that the mean deviation from the optimum error power is the same for both correlated and uncorrelated input samples and increases with the signal variance, which may lead to a degraded prediction gain of the RLS algorithm for high resolution audio signals. For real audio signals quantized at 16, 20 and 24 bits, simulation results confirm our theoretical analysis.

In this paper, an efficient implementation of the forward and inverse MDCT is proposed for even-length MDCT. The algorithm uses discrete cosine transform of type II (DCT-II) to compute the forward MDCT and their inverse DCT-III to compute the inverse MDCT. The lifting scheme is used to approximate multiplications appearing in the MDCT lattice structure where the dynamic range of the lifting coefficients can be controlled by proper choices of the lifting factorizations. The new structure requires less multiplications and additions than previous reported algorithms. The new transform has the properties that it is an integer-to-integer mapping and is reversible. Moreover, it inherits most of the attractive properties of the MDCT, including a good spectral representation of the audio signal, critical sampling and overlapping of blocks. Hence, the intMDCT is suitable for both lossless and lossy audio coding.

In this paper, a system using soft decision decoding of convolutional codes is investigated for MPEG-4 advanced audio coding. Under severe channel random bit error rates of 4.00x10/sup -2/ and above, the proposed scheme is able to achieve above 90% improvement in residual bit error rate performance over the current scheme. The average percentage overhead incurred using the proposed scheme is about 21%, which is the same as that of the original UEP scheme. An analysis with the theoretical bounds shows that within experimental error, the proposed scheme can perform to near optimum as compared to the current scheme.

This paper compares the eigenstructure and modulation algorithms, which are used for two-channel lossless FIR filter optimization. We study the effects of eigenvalue separation of the input covariance matrix and the step size on their convergence behavior. First, we show that the convergence rate of two algorithms increases as the separation of eigenvalues of the covariance matrix increases. The modulation algorithm converges more rapidly than the eigenstructure one because of its better eigenvalue separation. Second, the necessary condition for which the two algorithms converge is derived. Simulations are presented which support the analysis.

We consider the problem of adaptively optimizing a two-channel lossless FIR filter bank, which finds application in subband coding or wavelet signal analysis. Instead of using a gradient descent procedure-with its inherent problem of possible convergence to local minima-we consider two eigenstructure algorithms. Both algorithms feature a priori bounds on the output error variance at any convergent point, and based on simulations lead to solutions that lie acceptably close to a global minimum point of an output error cost function.