Audio Classification

In the field of audio classification, audio signals may be broadly divided into three classes: speech, music and events. Most studies, however, neglect that real audio soundtracks can have any combination of these classes simultaneously. In this study, a novel feature, “Entrocy”, is proposed for the detection of music both in pure form and overlapping with the other audio classes. Entrocy is defined as the variation of the information (or entropy) in an audio segment over time. Segments which contain music were found to have lower Entrocy since there are fewer abrupt changes over time.We have also compared Entrocy with existing music detection features and the entrocy showing a promising performance.Keywords—Music detection, audio content analysis, audio indexing, Entropy, real world audio classification.

In this work, we combine a Siamese neural network and different clustering techniques to generate a dissimilarity space that is then used to train an SVM for automated animal audio classification. The animal audio datasets used are (i) birds and (ii) cat sounds, which are freely available. We exploit different clustering methods to reduce the spectrograms in the dataset to a number of centroids that are used to generate the dissimilarity space through the Siamese network. Once computed, we use the dissimilarity space to generate a vector space representation of each pattern, which is then fed into an support vector machine (SVM) to classify a spectrogram by its dissimilarity vector. Our study shows that the proposed approach based on dissimilarity space performs well on both classification problems without ad-hoc optimization of the clustering methods. Moreover, results show that the fusion of CNN-based approaches applied to the animal audio classification problem works better than ...

This work addresses the recurring challenge of real-time monophonic and polyphonic audio source classification. The whole normalized power spectrum (NPS) is directly involved in the proposed process, avoiding complex and hazardous traditional feature extraction. It is also a natural candidate for polyphonic events thanks to its additive property in such cases. The classification task is performed through a nonparametric kernel-based generative modeling of the power spectrum. Advantage of this model is twofold: it is almost hypothesis free and it allows to straightforwardly obtain the maximum a posteriori classification rule of online signals. Moreover it makes use of the monophonic dataset to build the polyphonic one. Then, to reach the real-time target, the complexity of the method can be tuned by using a standard hierarchical clustering preprocessing of the prototypes, revealing a particularly efficient computation time and classification accuracy trade-off. The proposed method, called RARE (for Real-time Audio Recognition Engine) reveals encouraging results both in monophonic and polyphonic classification tasks on benchmark and owned datasets, including also the targeted real-time situation. In particular, this method benefits from several advantages compared to the state-of-the-art methods including a reduced training time, no feature extraction, the ability to control the computation -accuracy trade-off and no training on already mixed sounds for polyphonic classification.

In this work, we combine a Siamese neural network and different clustering techniques to generate a dissimilarity space that is then used to train an SVM for automated animal audio classification. The animal audio datasets used are (i) birds and (ii) cat sounds, which are freely available. We exploit different clustering methods to reduce the spectrograms in the dataset to a number of centroids that are used to generate the dissimilarity space through the Siamese network. Once computed, we use the dissimilarity space to generate a vector space representation of each pattern, which is then fed into an support vector machine (SVM) to classify a spectrogram by its dissimilarity vector. Our study shows that the proposed approach based on dissimilarity space performs well on both classification problems without ad-hoc optimization of the clustering methods. Moreover, results show that the fusion of CNN-based approaches applied to the animal audio classification problem works better than the stand-alone CNNs.

A real-time pitch modification system has been developed. The implemented processing scheme is based on hybrid deterministic/stochastic decomposition of the signal and includes extraction of instantaneous pitch, pitch-synchronous time-frequency analysis, parametrical morphing and synthesis. The scheme provides high quality output with considerably high naturalness. The aim of the presentation is to show capabilities of the designed real-time signal processing framework. The system implements speech-specific intonation change routines such as lowering, uplifting, tremolo etc. In order to make the presentation more expressive we designed a special singing mode in which the system automatically corrects wrong notes. The target melody and voice effects are specified using musical instruments digital interface (MIDI).

Audio classification is paramount in a variety of applications including surveillance, healthcare monitoring, and environmental analysis. Traditional methods frequently depend on intricate signal processing algorithms and manually crafted features, which may fall short in fully capturing the complexities of audio patterns. Neuromorphic computing, inspired by the architecture and functioning of the human brain, presents a promising alternative for audio classification tasks. This survey provides an exhaustive examination of the current state-ofthe-art in neuromorphic-based audio classification. It delves into the crucial components of neuromorphic systems, such as Spiking Neural Networks (SNNs), memristors, and neuromorphic hardware platforms, highlighting their advantages in audio classification. Furthermore, the survey explores various methodologies and strategies employed in neuromorphic audio classification, including event-based processing, spike-based learning, and bioinspired feature extraction. It examines how these approaches address the limitations of traditional audio classification methods, particularly in terms of energy efficiency, real-time processing, and robustness to environmental noise. Additionally, the paper conducts a comparative analysis of different neuromorphic audio classification models and benchmarks, evaluating their performance metrics, computational efficiency, and scalability. By providing a comprehensive guide for researchers, engineers and practitioners, this survey aims to stimulate further innovation and advancements in the evolving field of neuromorphic audio classification.

Hearing a species in a tropical rainforest is much easier than seeing them. If someone is in the forest, he might not be able to look around and see every type of bird and frog that are there but they can be heard. A forest ranger might know what to do in these situations and he/she might be an expert in recognizing the different type of insects and dangerous species that are out there in the forest but if a common person travels to a rain forest for an adventure, he might not even know how to recognize these species, let alone taking suitable action against them. In this work, a model is built that can take audio signal as input, perform intelligent signal processing for extracting features and patterns, and output which type of species is present in the audio signal. The model works end to end and can work on raw input and a pipeline is also created to perform all the preprocessing steps on the raw input. In this work, different types of neural network architectures based on Long Short Term Memory (LSTM) and Convolution Neural Network (CNN) are tested. Both are showing reliable performance, CNN shows an accuracy of 95.62% and Log Loss of 0.21 while LSTM shows an accuracy of 93.12% and Log Loss of 0.17. Based on these results, it is shown that CNN performs better than LSTM in terms of accuracy while LSTM performs better than CNN in terms of Log Loss. Further, both of these models are combined to achieve high accuracy and low Log Loss. A combination of both LSTM and CNN shows an accuracy of 97.12% and a Log Loss of 0.16.

Audio classification is a process of assigning particular class to an audio signal. Classifying the audio signal has many applications in the field of digital library, automatic organization of databases etc. In the last several years efforts have been made to develop different methods to extract information from audio signals, so that they may be stored, organized and retrieved automatically whenever required. In this work, audio signals are classified into different categories based on spectral and temporal features. In this methodology, the audio signal is initially decomposed into overlapped frames. Ensemble Empirical Mode Decomposition (EEMD), which is noise assisted data analysis method, is used to convert these frames into a set of band-limited functions known as Intrinsic Mode Functions (IMFs). Temporal and Spectral features then extracted from these IMFs and thereafter classification is done using Gaussian Mixture Model (GMM) classifier. Different combinations of features were tested to create feature vector. The experimental results showed accuracy of more than 80%. KEYWORDS--Intrinsic mode function, ensemble empirical mode decomposition, spectral and temporal features, Gaussian Mixture Model.

This work is devoted to the problem of automatic speech and music discrimination. As we will see here, speech and music signals have quite distinctive features. However, the efficient distinction between speech and music is still an open problem. This problem arises when it is necessary to extract speech information from the data containing both speech and music. The typical use of such segmentation is the extraction of speech segments from broadcast news for further processing by an Automatic Speech Recognition System (ASR). This work proposes a simple and quite effective solution to this problem based on the analysis of the speech and music modulation spectrum. The organisation of this paper is as follows. Section 2 discusses the main problems and possible solutions for the speech and music discrimination problem. Section 3 presents some results from the study of human speech perception that have been used in this work. Section 4 outlines our approach for music and speech discrimination. Experiment results of the method proposed here and result analysis are presented in Section 5. Finally, the conclusion will be given in Section 6. A short description of function implementing approach proposed here is presented in Appendix 1.

Speech Signals are the primary source of direct transmitter-to-receiver human communication and falls in the category of acoustic signals. These signals are the mechanical waves represented in terms of analog signal and propagate as vibration in the channel. Only one thing that classifies the speech signals in acoustic signals is its origination by humans. The signals possess most diverse features as characteristics of human development and culture. Henceforth, the applications of speech signals range from music, medicine to security, authentication etc. The transmission of speech signals commence at generation, propagation and conclude by reception of signals (figure 1.1).The unwanted signals (figure 1.2) from different sources mix with these signals and generate noise. The noise in digital communication can surface due to various reasons for example: finite precision of equipments or use of local components, faulty lines, improper coding etc. This type of noise is a different subj...

The sequence of strings played on a bowed string instrument is essential to understanding of the fingering. Thus, its estimation is required for machine understanding of violin playing. Audio-based identification is the only viable way to realize this goal for existing music recordings. A naïve implementation using audio classification alone, however, is inaccurate and is not robust against variations in string or instruments. We develop a bowed string sequence estimation method by combining audio-based bowed string classification and context-dependent error correction. The robustness against different setups of instruments improves by normalizing the F0-dependent features using the average feature of a recording. The performance of error correction is evaluated using an electric violin with two different brands of strings and and an acoustic violin. By incorporating mean normalization, the recognition error of recognition accuracy due to changing the string alleviates by 8 points, and that due to change of instrument by 12 points. Error correction decreases the error due to change of string by 8 points and that due to different instrument by 9 points.

In the age of digital information, audio data has become an important part in many modern computer applications. Audio classification has been becoming a focus in the research of audio processing and pattern recognition. Automatic audio classification is very useful to audio indexing, content-based audio retrieval and on-line audio distribution, but it is a challenge to extract the most common and salient themes from unstructured raw audio data. In this paper, we propose effective algorithms to automatically classify audio clips into one of six classes: music, news, sports, advertisement, cartoon and movie. For these categories a number of acoustic features that include linear predictive coefficients, linear predictive cepstral coefficients and mel-frequency cepstral coefficients are extracted to characterize the audio content. Support vector machines are applied to classify audio into their respective classes by learning from training data. Then the proposed method extends the application of neural network (RBFNN) for the classification of audio. RBFNN enables nonlinear transformation followed by linear transformation to achieve a higher dimension in the hidden space. The experiments on different genres of the various categories illustrate the results of classification are significant and effective.

It is both challenging and desirable to be able to retrieve sound files relevant to users' interests by searching the Internet. Unlike the traditional way of using keywords as input to search for web pages with relevant texts, query example can be used as input to search for similar sound files. In this paper, content-based technology has been applied to automatically retrieve sounds similar to the query-example. Features from time, frequency and coefficients domains are firstly extracted from each sound file. Next, Euclidean distances between the vectors of query and sample audios are measured. An ascending distance list is given as retrieval results. Experiments have been conducted on a sound database with 414 files from 16 classes. Further, we propose to classify the query audio into three classes, speech, music and other sound, with much fewer features and then search relevant files only in that subspace. This way, the retrieval performance could be further increased with the saving of computing time as well. Simulations show that our method leads to better results compared to the Soundfisher software in terms of both retrieval quality and completeness.

Speech Emotion Recognition (SER) is a method where computers learn to recognize human emotions from speech to improve communication. In this study, we present an innovative Bangla SER framework, incorporating data augmentations, feature extractions, and a deep learning model. We utilize a publicly available Bangla SER dataset named SUBESCO, developed by Shahjalal University of Science and Technology (SUST), which stands as one of the largest datasets in the Bangla SER domain. To enhance the data’s amount and variety, we use noise injection, time stretching, time shifting, and pitch augmentation. Previous Bangla SER models have struggled with severe misclassification in lower tonal classes, highlighting the need for more robust feature extraction. Therefore, we have conducted feature extraction, including Zero Crossing Rate (ZCR), Chroma stft, Melfrequency Cepstrum Coefficient (MFCC), Root Mean Square, and Mel Spectrogram, to improve classification accuracy. This paper aims to use a Deep Neural Network (DNN) for recognizing emotion labels from Bangla spoken speech data and has achieved 99% accuracy on unseen data. Given the scarcity of Bangla SER datasets and to validate the model’s robustness, we tested its performance on additional lingual datasets, namely RAVDESS, TESS, EMOVO, EmoDB, and BanglaSER. Remarkably, our model demonstrated high accuracy rates of 90%, 99%, 96%, 95%, and 93% on these datasets, respectively.

In recent years, deep learning approaches have gained significant interest as a way of building hierarchical representations from unlabeled data. However, to our knowledge, these deep learning approaches have not been extensively studied for auditory data. In this paper, we apply convolutional deep belief networks to audio data and empirically evaluate them on various audio classification tasks. In the case of speech data, we show that the learned features correspond to phones/phonemes. In addition, our feature representations learned from unlabeled audio data show very good performance for multiple audio classification tasks. We hope that this paper will inspire more research on deep learning approaches applied to a wide range of audio recognition tasks.

Many different short-time features, using time windows in the size of 10-30 ms, have been proposed for music segmentation, retrieval and genre classification. However, often the available time frame of the music to make the actual decision or comparison (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration and late information fusion 1 for music genre classification. A new feature integration technique, the AR model, is proposed and seemingly outperforms the commonly used meanvariance features.

This paper presents a method to classify audio-video data into one of five classes: advertisement, cartoon, news, movie and songs. Automatic audio-video classification is very useful to audio-video indexing, content based audio-video retrieval. Mel frequency cepstral coefficients are used to characterize the audio data. The color histogram features extracted from the images in the video clips are used as visual features. The experiments on different genres illustrate the results of classification are significant and effective. Experimental results of audio classification and video classification are combined using weighted sum rule for audio-video based classification. The method SVM and AANN classifies the audio-video clips with an accuracy of 95.54%., and 92.94%.

To assist clinicians in the differential diagnosis and treatment of motor speech disorders, it is imperative to establish objective tools which can reliably characterize different subtypes of disorders such as apraxia of speech (AoS) and dysarthria. Objective tools in the context of speech disorders typically rely on thousands of acoustic features, which raises the risk of difficulties in the interpretation of the underlying mechanisms, overadaptation to training data, and weak generalization capabilities to test data. Seeking to use a small number of acoustic features and motivated by the clinical-perceptual signs used for the differential diagnosis of AoS and dysarthria, we propose to characterize differences between AoS and dysarthria using only six handcrafted acoustic features, with three features reflecting segmental distortions, two features reflecting loudness and hypernasality, and one feature reflecting syllabification. These three different sets of features are used to separately train three classifiers. At test time, the decisions of the three classifiers are combined through a simple majority voting scheme. Preliminary results show that the proposed approach achieves a discrimination accuracy of 90%, outperforming using state-of-the-art features such as openSMILE which yield a discrimination accuracy of 65%.

In this paper, we describe an efficient method for audio matching which performs effectively for a wide range of classical music. The basic goal of audio matching can be described as follows: consider an audio database containing several CD recordings for one and the same piece of music interpreted by various musicians. Then, given a short query audio clip of one interpretation, the goal is to automatically retrieve the corresponding excerpts from the other interpretations. To solve this problem, we introduce a new type of chroma-based audio feature that strongly correlates to the harmonic progression of the audio signal. Our feature shows a high degree of robustness to variations in parameters such as dynamics, timbre, articulation, and local tempo deviations. As another contribution, we describe a robust matching procedure, which allows to handle global tempo variations. Finally, we give a detailed account on our experiments, which have been carried out on a database of more than 110 hours of audio comprising a wide range of classical music.

Modern streaming services are increasingly labeling videos based on their visual or audio content. This typically augments the use of technologies such as AI and ML by allowing to use natural speech for searching by keywords and video descriptions. Prior research has successfully provided a number of solutions for speech to text, in the case of a human speech, but this article aims to investigate possible solutions to retrieve sound events based on a natural language query, and estimate how effective and accurate they are. In this study, we specifically focus on the YamNet, AlexNet, and ResNet-50 pre-trained models to automatically classify audio samples using their respective melspectrograms into a number of predefined classes. The predefined classes can represent sounds associated with actions within a video fragment. Two tests are conducted to evaluate the performance of the models on two separate problems: audio classification and intervals retrieval based on a natural language query. Results show that the benchmarked models are comparable in terms of performance, with YamNet slightly outperforming the other two models. YamNet was able to classify single fixed-size audio samples with 92.7% accuracy and 68.75% precision while its average accuracy on intervals retrieval was 71.62% and precision was 41.95%. The investigated method may be embedded into an automated event marking architecture for streaming services.

Modern streaming services are increasingly labeling videos based on their visual or audio content. This typically augments the use of technologies such as AI and ML by allowing to use natural speech for searching by keywords and video descriptions. Prior research has successfully provided a number of solutions for speech to text, in the case of a human speech, but this article aims to investigate possible solutions to retrieve sound events based on a natural language query, and estimate how effective and accurate they are. In this study, we specifically focus on the YamNet, AlexNet, and ResNet-50 pre-trained models to automatically classify audio samples using their respective melspectrograms into a number of predefined classes. The predefined classes can represent sounds associated with actions within a video fragment. Two tests are conducted to evaluate the performance of the models on two separate problems: audio classification and intervals retrieval based on a natural language query. Results show that the benchmarked models are comparable in terms of performance, with YamNet slightly outperforming the other two models. YamNet was able to classify single fixed-size audio samples with 92.7% accuracy and 68.75% precision while its average accuracy on intervals retrieval was 71.62% and precision was 41.95%. The investigated method may be embedded into an automated event marking architecture for streaming services.

Automatic discrimination of speech and music is an important tool in many multimedia applications. Previous work has focused on using long-term features such as differential parameters, variances, and time-averages of spectral parameters. These classifiers use features estimated over windows of 0.5-5 seconds, and are relatively complex. In this paper, we present our results of combining the line spectral frequencies (LSFs) and zero-crossing-based features for frame-level narrowband speech/music discrimination. Our classification results for different types of music and speech show the good discriminating power of these features. Our classification algorithms operate using only a frame delay of 20 ms, making them suitable for real-time multimedia applications.

In recent years, deep learning approaches have gained significant interest as a way of building hierarchical representations from unlabeled data. However, to our knowledge, these deep learning approaches have not been extensively studied for auditory data. In this paper, we apply convolutional deep belief networks to audio data and empirically evaluate them on various audio classification tasks. In the case of speech data, we show that the learned features correspond to phones/phonemes. In addition, our feature representations learned from unlabeled audio data show very good performance for multiple audio classification tasks. We hope that this paper will inspire more research on deep learning approaches applied to a wide range of audio recognition tasks.

Speech and Audio processing techniques are used along with statistical pattern recognition principles to solve the problem of music instrument recognition. Non temporal, frame level features only are used so that the proposed system is scalable from the isolated notes to the solo instrumental phrases scenario without the need for temporal segmentation of solo music. Based on their effectiveness in speech, Line Spectral Frequencies(LSF) are proposed as features for music instrument recognition. The proposed system has also been evaluated using MFCC and LPCC features. Gaussian Mixture Models and K-Nearest Neighbour model classifier are used for classification. The experimental dataset included the UIowa's MIS and the C Music corporation's RWC databases. Our best results at the instrument family level is about 95% and at the instrument level is about 90% when classifying 14 instruments.

This report describes the use of a support vector machines with a novel kernel, to determine the breathing rate and inhalation duration of a fire fighter wearing a Self-Contained Breathing Apparatus. With this information, an incident commander can monitor the firemen in his command for exhaustion and ensure timely rotation of personnel to ensure overall fire fighter safety

Les services audio de la nouvelle génération requièrent des outils d'édition des fichiers audio qui permettent de traiter un tel fichier aussi simplement qu'un fichier texte. L'indexation des fichiers est une solution permettant l'accès rapideà l'information qui peutêtre l'identité d'un locuteur, la localisation de son intervention dans le temps, le sujet de la conversation. D'excellents outils de traitement de texte existent depuis de nombreuses années et une solution pour la navigation aisée dans un fichier audio pourrait se réduireà la reconnaissance du contenu entier du fichieràéditer (parole vers texte) mais requiert en général des reconnaisseurs de grands vocabulaires et indépendants des locuteurs qui fournissent actuellement des résultats acceptables seulement pour des locuteurs coopératifs limitant le thème de leur intervention au domaine applicatif correspondantà un modèle de langage prédéfini. Et même dans ce cas, la détection des interludes musicaux, l'identification d'un intervenant et la segmentation de l'enregistrement en zones correspondantà un seul locuteur restent des problèmes d'intérêt. La maitrise complète des techniques d'indexation ouvrira un grand marché d'applications attrayantes pour le consommateur telles que la production de programmes audio (et aussi video)à la demande mais aussi pour le monde professionnel de la radiophonie qui pourrait composer du contenu par réutilisation. Il tombe aussi sous le sens que de tels outils faciliteront l'accès des bases de données et des archives multimedia.

Script identification is challenging task in bilingual or multilingual optical character recognition system. A remarkable research work on script identification have been noted in Indian or non-Indian context. As many commercial and official regional documents of different states of India are in bilingual containing one regional language of respective state and the other international intersperse language English. Therefore script identification is one of the primary tasks in multi-script document recognition. English words are mostly interspersed in regional documents of different states of India. In this paper script identification of Gujarati and English at word level is presented. For feature extraction the directional energy distribution of a word using Gabor filters is used with suitable frequencies and orientations. The proposed system uses SVM classifier to classify the extracted features in one of the script. The results obtained are quiet encouraging.

Contemporary audio declipping algorithms often ignore the possibility of the presence of additive channel noise. If and when noise is present, however, the efficacy of any declipping algorithm is critically dependent on the accuracy with which clipped portions of the signal can be detected. This paper introduces an effective technique for inferring the amplitude and percentile values of the clipping threshold, and develops a statistically-optimal classification algorithm for accurately differentiating between clipped and unclipped samples in a noisy speech signal. The overall effectiveness of the clipped sample estimation algorithm is evaluated by the degree to which automatic speech recognition performance is improved when decoding clipped speech that has been declipped with state-of-the-art declipping algorithms paired with the clipped sample estimation algorithm. Up to 35% relative improvements in word error rate have been observed. Beyond the accuracy of the developed techniques, this paper generally underscores the necessity of robust parameter estimation methods for declipping in noise.

In this paper we present a study on music mood classification using audio and lyrics information. The mood of a song is expressed by means of musical features but a relevant part also seems to be conveyed by the lyrics. We evaluate each factor independently and explore the possibility to combine both, using Natural Language Processing and Music Information Retrieval techniques. We show that standard distance-based methods and Latent Semantic Analysis are able to classify the lyrics significantly better than random, but the performance is still quite inferior to that of audio-based techniques. We then introduce a method based on differences between language models that gives performances closer to audio-based classifiers. Moreover, integrating this in a multimodal system (audio+text) allows an improvement in the overall performance. We demonstrate that lyrics and audio information are complementary, and can be combined to improve a classification system.

Due to age-bound onset of symptoms used for diagnosis of mild to moderate intellectual disability, early diagnosis of these problems has long been a difficult issue. The diagnosis includes tests pertaining to intellectual functioning and adaptive behaviours including communication skills etc. In this paper, it is proposed to use speech features as an early indicator of the disorder which can be used to train machine learning algorithms for differentiating between speech of normally developing children and children with intellectual disability. In this paper, speech abnormalities are quantified using acoustic parameters including Linear Predictive Cepstral Coefficients, Mel Frequency Cepstral Coefficients and spectral features in speech samples of 48 participants (24 with intellectual disability and 24 age-matched controls). A training dataset was created by extracting these features which was used for learning by various classifiers. The experiments show promising results where Support Vector Machine gives an accuracy of 98%. Consequently, a well-trained classification algorithm can be used as an aid in early detection of mild to moderate intellectual disability.

We study auditory context recognition for contextaware mobile computing systems. Auditory contexts are recordings of a mixture of sounds, or ambient audio, from mobile users' everyday environments. For training a classifier, a set of recordings from different environments are segmented and labeled. The segments are windowed into overlapping frames for feature extraction. While previous work in auditory context recognition has often treated the problem as a sequence classification task and used HMM-based classifiers to recognize a sequence of consecutive MFCCs of frames, we compute averaged Mel-spectrum over the segments and train a SVM-based classifier. Our scheme outperforms an already reported HMMbased scheme. This result is achieved using the same dataset. We also show that often the feature sets used by previous work are affected by attenuation, limiting their applicability in practice. Furthermore, we study the impact of segment duration on recognition accuracy.

Due to the progress of the unlimited data storage capabilities and the proliferation use of the Internet, information retrieval systems encountered a large interest. Much of this data is in different forms from various sources. So, it becomes important to develop the necessary technologies for indexing and browsing such audio data. The consideration will be from audio information retrieval domain. In this paper; the study of audio content analysis for classification is presented, in which an audio signal is classified according to audio type. An approach that is capable of classifying an audio signal into speech, music, environment sound is used. Audio classification is processed in two steps as follows:-The first step of the classification is speech and nonspeech discrimination. The second step further divides nonspeech class into music, background sounds. Algorithms based on K-nearest-neighbor (KNN) and Fast Fourier Transform (FFT) are used. Experimental results indicate that satisfactory results are produced.

In this paper, starting from a robust statistics (RS) adaptive approach presented in a previous work entitled the combined NLMS-Sign (CNLMS-S) adaptive filter, an automatic combination technique with similar performances is proposed. Thus, in order to obtain better performances in acoustic echo cancellation (AEC) setups than with the normalized least-mean square (NLMS) algorithm, in the CNLMS-S case the decision between the two algorithms (NLMS and Sign) is based on a set error threshold. The error threshold can be empirically determined or known a priori if the signal-to-noise ratio (SNR) value from the loudspeaker-enclosure-microphone (LEM) setup is available or if the local noise levels can be determined from the silences. Here, to overcome this shortcoming, an adaptive combination of the two algorithms involved in RS is highlighted, providing similar results regarding convergence and final misadjustment. Also, the need of the error threshold set by the user is removed, the combi...

Presently, fast proliferation of information enforces novel challenges on content management. Further, computerized audio classification along-with content description is considered as valuable method to manage audio contents. In general, classification involves two steps. First, is the processing of accessible data in economical ways to deliver explanatory features. Second is how accurate features of undetermined tests is evaluated to choose classifier. In this paper, k-neighbor algorithm with machine learning is proposed for feature extraction as well as content classification/description. This algorithm enhances Quality of Service parameters of classifiers. Here, development of training as well as testing data set is developed to increase the classifier accuracy. A test engine setup bed using simulation tool MATLAB is designed to estimate the implementation performance of the algorithm. A range of features are studied to evaluate effectiveness in terms of accuracy, zero crossing rate (ZCR) and spectral roll frequency. From the experimentation results, it is observed that the proposed algorithm can achieve accuracy of 95.8% for 2 sec window length as compare with kneighbor algorithm. A total enhancement of 11% is achieved with cross validation error of 29.6. A superior assortment of training fabric to extract few additional useful features can enhance accuracy further.

Many audio and multimedia applications would benefit if they could interpret the content of audio rather than relying on descriptions or keywords. These applications include multimedia databases and file systems, digital libraries, automatic segmentation or indexing of video (e.g., news or sports storage), and surveillance. This paper describes a novel content-based audio classification approach based on neural network and genetic algorithm. Experiments show this approach achieves a good performance of the classification.

Musical signals exhibit periodic temporal structure that create the sensation of rhythm. In order to model, analyze, and retrieve musical signals it is important to automatically extract rhythmic information. To somewhat simplify the problem, automatic algorithms typically only extract information about the main beat of the signal which can be loosely defined as the regular periodic sequence of pulses corresponding to where a human would tap his foot while listening to the music. In these algorithms, the beat is characterized by its frequency (tempo), phase (accent locations) and a confidence measure about its detection. The main focus of this paper is the concept of Beat Strength, which will be loosely defined as one rhythmic characteristic that could allow to discriminate between two pieces of music having the same tempo. Using this definition, we might say that a piece of Hard Rock has a higher beat strength than a piece of Classical Music at the same tempo. Characteristics related to Beat Strength have been implicitely used in automatic beat detection algorithms and shown to be as important as tempo information for music classification and retrieval. In the work presented in this paper, a user study exploring the perception of Beat Strength was conducted and the results were used to calibrate and explore automatic Beat Strength measures based on the calculation of Beat Histograms.

The barkhausen noise carries important information which can be used in early damage detection and fault diagnosis. The barkhausen noise is corrupted by interference signals from other sources during the measure and the information of fault can be lost. In this paper a new algorithm based on Empirical Mode Decomposition (EMD) and Savitzky-Golay filter is proposed to extract the information about the fault of materials from a measured barkhausen noise. Firstly, using EMD to decompose the barkhausen noise signal into elementary function called Intrinsic Mode Function (IMF). Secondly, we use the energy to select the relevant mode, these selected IMF is filtered by Savitzky-Golay filter and the reconstructed signal is obtained by the IMFs filtered. The envelope spectra are used to test the efficiency of the proposed method in enhancement the quality of barkhausen noise signal. The proposed method is validated using the measuring chain of barkhausen noise (Fig (3-8)) and the obtained res...

Acoustic monitoring is crucial for the conservation and management of ecosystems and their flora and fauna. Traditional animal monitoring is based on passive acoustic recordings that are manually assessed by human experts, which adds a significant workload. To alleviate this burden, Machine Learning (ML)-based solutions for automating this process have been in the spotlight.

This research explores end-to-end Deep Learning (DL) techniques for species recognition using bioacoustic data from the Internet of Things (IoT) devices installed in remote wild areas. The reason is twofold. First, DL-based techniques extract hierarchical features automatically through multiple layers of abstraction. Therefore, it eliminates the very expensive and time-consuming process of handcrafting the features required in Non-Deep Learning (Non-DL)-based techniques. Second, we try to leverage the resounding success of DL in producing state-of-the-art (SOTA) performance for a wide range of applications to acoustic analysis.

Due to the unavailability of power sources and the internet in remote areas, the extremely resource-constrained IoT devices (battery powered, low memory, and low processing power) cannot send the data to the cloud server for recognition. As a result, the recognition activities must be performed locally on IoT devices. However, DL-based techniques are resource and time-intensive. To harness the power of DL in IoT, the DL models must be significantly resource-optimized.

This study focuses on significantly downsizing the DL model architecture without sacrificing accuracy. It develops a compression-friendly Deep Convolutional Neural Network (Deep-CNN) architecture for end-to-end raw audio classification, a structured compression technique, and a generic pipeline that compresses this Deep-CNN architecture to produce a tiny DL model suitable for deployment on a resource-impoverished Microcontroller Unit (MCU). Beyond theorizing, we demonstrate such a deployment practically.

One of the major challenges in biological applications is the availability of labeled real-world data for the training phase. We demonstrate how Active Learning (AL) can be used with our edge models to address this issue. Our approach goes beyond standard AL by incorporating feature extraction into the AL loop, and we show that this results in superior performance.

To demonstrate the relevance of our work for real-world applications, we go beyond standard benchmarks and test the applicability of our methods with real-world data from conservation biology projects.

Deep learning has celebrated resounding successes in many application areas of relevance to the Internet of Things (IoT), such as computer vision and machine listening. These technologies must ultimately be brought directly to the edge to fully harness the power of deep leaning for the IoT. The obvious challenge is that deep learning techniques can only be implemented on strictly resource-constrained edge devices if the models are radically downsized. This task relies on different model compression techniques, such as network pruning, quantization, and the recent advancement of XNOR-Net. This study examines the suitability of these techniques for audio classification on microcontrollers. We present an application of XNOR-Net for end-to-end raw audio classification and a comprehensive empirical study comparing this approach with pruning-and-quantization methods. We show that raw audio classification with XNOR yields comparable performance to regular full precision networks for small numbers of classes while reducing memory requirements 32-fold and computation requirements 58-fold. However, as the number of classes increases significantly, performance degrades, and pruning-and-quantization based compression techniques take over as the preferred technique being able to satisfy the same space constraints but requiring approximately 8x more computation. We show that these insights are consistent between raw audio classification and image classification using standard benchmark sets. To the best of our knowledge, this is the first study to apply XNOR to end-to-end audio classification and evaluate it in the context of alternative techniques. All codes are publicly available on GitHub.

Significant effort s are being invested to bring state-of-the-art classification and recognition to edge devices with extreme resource constraints (memory, speed, and lack of GPU support). Here, we demonstrate the first deep network for acoustic recognition that is small, flexible and compression-friendly yet achieves state-of-the-art performance for raw audio classification. Rather than handcrafting a once-off solution, we present a generic pipeline that automatically converts a large deep convolutional network via compression and quantization into a network for resource-impoverished edge devices. After introducing ACDNet, which produces above state-of-the-art accuracy on ESC-10 (96.65%), ESC-50 (87.10%), Urban-Sound8K (84.45%) and AudioEvent (92.57%), we describe the compression pipeline and show that it allows us to achieve 97.22% size reduction and 97.28% FLOP reduction while maintaining close to state-of-the-art accuracy 96.25%, 83.65%, 78.27% and 89.69% on these datasets. We describe a successful implementation on a standard off-the-shelf microcontroller and, beyond laboratory benchmarks, report successful tests on real-world datasets.

A hybrid speech/non-speech detector is proposed for the pre-processing of broadcast news. During the first stage speech/non-speech classification of uniform overlapping segments is performed. The accuracy in the detection of boundaries is determined by the degree of overlap of the audio segments and it is 250 ms in our case. Extracted speech segments are further processed on a frame basis

In this work, we combine a Siamese neural network and different clustering techniques to generate a dissimilarity space that is then used to train an SVM for automated animal audio classification. The animal audio datasets used are (i) birds and (ii) cat sounds, which are freely available. We exploit different clustering methods to reduce the spectrograms in the dataset to a number of centroids that are used to generate the dissimilarity space through the Siamese network. Once computed, we use the dissimilarity space to generate a vector space representation of each pattern, which is then fed into an support vector machine (SVM) to classify a spectrogram by its dissimilarity vector. Our study shows that the proposed approach based on dissimilarity space performs well on both classification problems without ad-hoc optimization of the clustering methods. Moreover, results show that the fusion of CNN-based approaches applied to the animal audio classification problem works better than ...

In this paper, a novel collective network of binary classifiers (CNBC) framework is presented for content-based audio classification. The topic has been studied in several publications before, but in many cases the number of different classification categories is quite limited and needed to be fixed a priori. We focus our efforts to increase both the classification accuracy and the number of classes, as well as to create a scalable network design, which allows introducing new audio classes incrementally. The approach is based on dividing a major classification problem into several networks of binary classifiers (NBCs), where each NBC adapts its internal topology according to the classification problem at hand, by using evolutionary Artificial Neural Networks (ANNs). In the current work, feedforward ANNs, or the so-called Multilayer Perceptrons (MLPs), are evolved within an architecture space, where a stochastic optimization is applied to seek for the optimal classifier configuration and parameters. The performance evaluations of the proposed framework over an 8-class benchmark audio database demonstrate its scalability and notable potential, as classification error rates of less than 9% are achieved.

The recently proposed Progressive Query method is a dynamic retrieval technique, which is mainly designed to bring an effective solution especially for queries on large-scale multimedia databases and furthermore to provide periodic fractional query retrievals along with the ongoing query process. In this way it achieves a series of sub-query results that will eventually be converging to the full-scale search retrieval in a faster and with no minimum system requirements. Due to its pre-emptive design over a single process, the precision of ...

Support vector machines (SVMs) have been recently proposed as a new learning algorithm for pattern recognition. In this paper, the SVMs with a binary tree recognition strategy are used to tackle the audio classification problem. We illustrate the potential of SVMs on a common audio database, which consists of 409 sounds of 16 classes. We compare the SVMs based classification with other popular approaches. For audio retrieval, we propose a new metric, called distance-from-boundary (DFB). When a query audio is given, the system first finds a boundary inside which the query pattern is located. Then, all the audio patterns in the database are sorted by their distances to this boundary. All boundaries are learned by the SVMs and stored together with the audio database. Experimental comparisons for audio retrieval are presented to show the superiority of this novel metric to other similarity measures.

This abstract describes the audio feature extraction and classification algorithm used for the University of Victoria submission to the MIREX (Music Information Retrieval Exchange) 2005. The same audio features and classification algorithm are used for the audio genre classification and artist identification tasks and arge therefore combined in this abstract. The feature set used is similar to one described in Tzanetakis and Cook (2002) but not including the Beat and Pitch Histogram features. This decision was made in order to ...