A method for detecting whistles, moans, and other frequency contour sounds

An algorithm is presented for the detection of frequency contour sounds-whistles of dolphins and many other odontocetes, moans of baleen whales, chirps of birds, and numerous other animal and non-animal sounds. The algorithm works by tracking spectral peaks over time, grouping together peaks in successive time slices in a spectrogram if the peaks are sufficiently near in frequency and form a smooth contour over time. The algorithm has nine parameters, including the ones needed for spectrogram calculation and normalization. Finding optimal values for all of these parameters simultaneously requires a search of parameter space, and a grid search technique is described. The frequency contour detection method and parameter optimization technique are applied to the problem of detecting "boing" sounds of minke whales from near Hawaii. The test data set contained many humpback whale sounds in the frequency range of interest. Detection performance is quantified, and the method is found to work well at detecting boings, with a false-detection rate of 3% for the target missed-call rate of 25%. It has also worked well anecdotally for other marine and some terrestrial species, and could be applied to any species that produces a frequency contour, or to non-animal sounds as well.     

Sound production by North Atlantic right whales (Eubalaena glacialis) in surface active groups

The surface active group (SAG) is the most commonly observed surface social behavior of North Atlantic right whales. Recordings were made from 52 SAGs in the Bay of Fundy, Canada between July and September, from 1999 to 2002. The call types recorded from these groups were similar to those described previously for Southern right whales (Eubalaena australis), with six major call types being termed scream, gunshot, blow, upcall, warble, and downcall. The percentage of total calls of each call type depended on the group size and composition. The most common call type recorded was the scream call. The scream calls were produced by the focal female in a SAG. Production of other sound types can be attributed to whales other than the focal female, with gunshot and upcalls produced by males, and warble calls produced by female calves. The source levels for these sounds range from 137 to 162 dB rms re 1 ,tPa-m for tonal calls and 174 to 192 dB rms for broadband gunshot sounds.     

Real-time aircraft noise likeness detector

One of the most difficult tasks involved in the process of noise monitoring near airports is related to the automatic detection and classification of aircraft noise events.These tasks can be solved by applying pattern recognition techniques to the audio signal captured by a microphone. But now the problem is caused by the background noise, which is present in real environments.This paper proposes a real-time method for continuously tracking the similarity of the input sound and the aircraft’s sounds. Using these facilities, the monitoring unit will be able to mark aircraft events, or to make measurements only when aircraft sound is louder than background noise.A one-class approach has been applied to this detection-by-classification method.Using the default setup, 93% of the aircraft’s events which held an SNR of 6-8 dB were detected, for 30 different locations with diverse soundscapes.     

Multiband analysis and synthesis of spectro-temporal modulations of Fourier spectrogram

The two-dimensional spectro-temporal modulation filtering concept of the auditory model [T. Chi, P. Ru, and S. A. Shamma, J. Acoust. Soc. Am. 118(2), 887-906 (2005)] is implemented on the Fourier spectrogram. The Fourier magnitude spectrogram is analyzed in terms of its joint spectro-temporal modulations, which embed the temporal dynamics and spectral structures. Instead of iterative projection methods, the overlap-and-add method is adopted to invert modified Fourier spectrograms back to sounds. The proposed framework not only provides a similar spectro-temporal analytical process for sounds as the auditory model but also produces synthesized sounds with better quality in a timely manner, which makes proposed framework feasible to human speech recognition (HSR) applications as well.     

Wide-band dereverberation method based on multichannel linear prediction using prewhitening filter

Several dereverberation algorithms have been studied. The sampling frequencies used in conventional studies are typically 8–16 kHz because their main purpose is preprocessing for improving the intelligibility of speech communication and articulation for automatic speech recognition. However, in next-generation communication systems, techniques to analyze and reproduce not only semantic information of sound but also more high-definition components such as spatial information and directivity will be increasingly necessary. To decompose these sound field characteristics with high definition, a dereverberation algorithm that is useful at high sampling frequencies is an important technique to process sound that includes high-frequency spectra such as musical sounds. The LInear-predictive Multichannel Equalization (LIME) algorithm is a promising dereverberation method. Using the LIME algorithm, however, a dereverberation signal cannot be solved at high sampling frequencies when the source signal is colored, such as in the case of speech and sound of musical signals. Because the rank of the correlation matrix calculated from such a colored signal is not full, the characteristic polynomial cannot be calculated precisely. To alleviate this problem, we propose preprocessing of all input signals with filters to whiten their spectra so that this algorithm can function for colored signals at high sampling frequencies.     

Polyspectral technique for the analysis of stress-waves characteristics and species recognition in wood veneers

In this work a simple technique to obtain information about the species of wood samples using stress-wave sounds in the audible range is presented. However, spectra of wood sounds generated by pendulum impact are very complex and feature extraction for classification purposes is very difficult. Polyspectral techniques have been successfully applied to several problems from radar pattern recognition to medical signal processing. Following this approach, convolution of four different sound impacts has been done. This permits to extract clear polyspectral features suitable for wood species recognition with possible applications to both human assisted and automatic wood identification systems with minimal intersample variability. Results indicate that using this technique only the two most intense polyspectral peaks are enough for species recognition.     

声纹的变压器放电与机械故障诊断研究*

变压器中的一些放电和机械故障会产生异常声音,可用于故障检测。据此,本文提出基于可听声的变压器放电和机械故障诊断方法。针对机械故障声音与变压器本体噪声特征相似易混淆的问题提出改进小波包-BP神经网络算法,与传统小波包-BP神经网络算法相比声音的识别率提高了5.7%。为提高声音识别系统的泛化性,提出基于梅尔对数频谱和卷积神经网络的声音识别算法。两种算法相互验证,提高了系统的可靠性。在真实变压器油箱中模拟了不同类型放电和机械故障。试验结果表明,本文提出的两种方法能成功识别放电和机械故障的声音,声音识别率分别为99.6％和97.57％。     

Analysis of pain-related vocalization in young pigs

The assessment of pain constitutes a major issue for animal welfare research. The objective of this study was to classify vocalizations during castration pain and to assess alterations in vocalizations under local anaesthesia. The alterations in vocalization were measured by multiparametric call analysis. A total of 4537 calls of 70 young pigs were evaluated. With the data of this study three call types are distinguishable (grunt, squeal, scream). A high percentage (94.64%) of calls that could be classified in one of the three call types during the castration process within the confidence level of 95% was found. The comparison of the occurrence of the call types during treatments gives evidence for pain-related use of screams. The piglets castrated without local anaesthesia produced almost double the number of screams as piglets castrated with anaesthesia. The comparison of the recorded sound parameters reveals the particular position of screams in the call repertoire of young pigs. Screams are significantly different in their sound parameters than grunts or squeals. Castration in comparison to mere restraint produced a comprehensive change in sound parameters, with castration calls becoming more extended and more powerful. The findings in this study also show differences in the effectiveness of the parameters which indicate pain. Parameters that describe a single event in a call, such as peak level or peak frequency give better results than parameters that describe an average, such as weighted frequency and main frequency. The research indicated that pain-related changes of calls in piglets can be identified. On the basis of the results, automatic classification of call types during management operations may be developed. This could contribute to objective animal welfare assessment.     

利用冲击声辨识声源材料的特征提取

声源辨识属于环境声识别的范畴,是模式识别的一个重要研究方向。冲击声携带了大量的声源物理信息,因此利用冲击声提取特征进行声源材料辨识是提高声目标识别分类性能的重要途径。对球-板撞击物理模型合成的冲击声连续统,提出使用基函数学习法提取目标特征,同时利用短时傅里叶变换和小波包变换进行特征提取,以此为基础完成被击平板的材料识别。研究结果表明,利用基函数学习法获得的特征,对于冲击声分类的效果明显优于短时傅里叶变换和小波包变换方法,说明利用该方法进行冲击声声源材料辨识的可行性和优势。      

The influence of signal parameters on the sound source localization ability of a harbor porpoise (Phocoena phocoena)

It is unclear how well harbor porpoises can locate sound sources, and thus can locate acoustic alarms on gillnets. Therefore the ability of a porpoise to determine the location of a sound source was determined. The animal was trained to indicate the active one of 16 transducers in a 16-m-diam circle around a central listening station. The duration and received level of the narrowband frequency-modulated signals (center frequencies 16, 64 and 100 kHz) were varied. The animal's localization performance increased when the signal duration increased from 600 to 1000 ms. The lower the received sound pressure level (SPL) of the signal, the harder the animal found it to localize the sound source. When pulse duration was long enough (approximately 1 s) and the received SPLs of the sounds were high (34-50 dB above basic hearing thresholds or 3-15 dB above the theoretical masked detection threshold in the ambient noise condition of the present study), the animal could locate sounds of the three frequencies almost equally well. The porpoise was able to locate sound sources up to 124 degrees to its left or right more easily than sounds from behind it.     

Feature extraction for sound source material recognition with impact sounds

Sound source recognition is a part of environmental sound recognition,which is one of the most important research areas in pattern recognition.Impact sounds carry much physical information associated with the sound sources,which makes impact sound based sound source recognition an important approach to improve recognition performance.In this study,the impact sound continuum synthesized with a ball-plate collision model is used for material recognition of the impacted plates.The basis function learning method and time-frequency representation methods,including the short time Fourier transform and the wavelet packet transform,are applied into classification and the recognition results are compared.The result shows that the features obtained by using the basis function learning perform better for material classification of the impacted plates than that by using the short time Fourier transform and the wavelet packet transform.This demonstrates the high efficiency and superiority of this method in material recognition of sound sources.     

Detection of extraneous abnormal sounds affecting road traffic noise by use of a necessary condition method

When measuring and/or recording road traffic sound levels during a long time interval, extraneous abnormal sounds will inevitably affect the road traffic sound levels of interest. Such sounds include those produced by horns, sirens, animals, construction sites, and the like. The detection and elimination of such extraneous sound requires much time and effort, but are necessary if noise indices such as L_eq, L_max, and L₁₀ are to be properly estimated. This paper proposes a practical detection method of these extraneous interfering sounds by deriving a necessary condition that road traffic sound levels must satisfy. The necessary condition provides an easy method of identifying sound levels not satisfying the condition, and distinguishes them as extraneous abnormal sounds, even in a large volume of observed data. The validity and usefulness of this method are confirmed by application to actually observed data.     

基于桥接Transformer的小样本优化鸟声识别网络

针对实际鸟类监测环境中,收集鸟声声频数据分布不均匀,导致神经网络训练不充分,分类识别测试准确率低的问题,设计了一种桥接Transformer神经网络模型。该网络首先利用原始鸟声声频信号生成短时傅里叶变换语谱图作为输入特征,之后将语谱图输入到由注意力模块和卷积模块桥接组成的Transformer网络中,完成对语谱图中全局特征和局部特征的信息交互,最后利用单层Transformer编码器实现对每一个批次样本的损失优化,得到最终的分类结果。在Birdsdata和xeno-canto鸟声数据集上进行小样本实验,分别获得了91.34%和82.63%的平均准确率,与其他鸟声识别网络进行了对比实验,验证了该网络的有效性。     

用图样匹配法在计算机上自动识别语音

本文中提出了一个机器识别语言的模型。根据这个模型,提出一种将图样在时间域上规正的新方法。最后给出一组实验结果,正确识别率超过99％。这表明,该方案能应用于某些自动控制系统中。     

贝叶斯优化卷积神经网络公共场所异常声识别*

针对公共场所异常声的感知和识别问题,提出一种基于贝叶斯优化卷积神经网络的识别方法。提取声信号的Gammatone倒谱系数、倍频程功率谱、短时能量和谱质心,组合成声信号的特征图。构建卷积神经网络作为分类器,利用递增的卷积核设置和池化操作处理不同尺度的特征。基于贝叶斯优化算法优化卷积神经网络的模型参数,对包括火苗噼啪声、婴儿啼哭声、烟花燃放声、玻璃破碎声和警报声的5种公共场所异常声进行识别。该方法的识别结果与基于不同的特征提取和分类器方案得到的识别结果进行比较,结果表明该方法的识别效果优于其他特征提取和分类器方案的识别效果。最后分析了该方法在不同信噪比噪声干扰下的识别结果,验证了该方法的有效性。     

Les methodes de determination de la sonie des bruits

For some considerable time now specialists have taken an interest in the measurement and analysis of noise. They quickly understood that the process of perception of sounds by the ear was very complex and that it was difficult to relate a physical level or a sound level to people's impression of nuisance or loudness. Therefore several authors such as, for example, Stevens, Kryter and Zwicker, proposed different methods and different units. In a general way, these methods apply to continuous noise but they cannot be applied without care to discontinuous impulsive or transitory noise. Speech and music come into this latter category. The known methods according to the present author cannot be used for dealing with this type of noise.     

An acoustic system for the individual recognition of insects

Research into acoustic recognition systems for insects has focused on species identification rather than individual identification. In this paper, the feasibility of applying pattern recognition techniques to construct an acoustic system capable of automatic individual recognition for insects is investigated analytically and experimentally across two species of Orthoptera. Mel-frequency cepstral coefficients serve as the acoustic feature, and α-Gaussian mixture models were selected as the classification models. The performance of the proposed acoustic system is promising and displays high accuracy. The results suggest that the acoustic feature and classifier method developed here have potential for individual animal recognition and can be applied to other species of interest.     

Sound characteristics of big-snout croaker, Johnius macrorhynus (Sciaenidae)

The sounds of a big-snout croaker, Johnius macrorhynus, produced under hand-held and voluntary conditions (in a large aquarium and in the field) were compared. Voluntary calls included "purr" and "dual-knocks", only purrs were produced when the fish was hand-held. The purr is composed of pulses in which the first interpulse interval was six to nine times longer than the other interpulse intervals, which were approximate in duration, and is a unique sound type in the coastal water of Taiwan. Purrs emitted under these conditions did not differ significantly, suggesting that the hand-held sound can be employed to match the sound in the field. These sounds contained energy reaching 5 kHz, with two peaks at about 1 and 2 kHz-the former being the dominant frequency. First interpulse interval, main interpulse interval, repetition rate of pulse, and pulse duration may serve as the diagnostic characters for the species-specific sound (i.e., purrs). One specimen survived in a large aquarium for 6 months and its vocal activity was monitored. The fish produced fewer dual-knocks than purrs, and purrs were about 11 dB louder than dual-knocks emitted in the aquarium. The temporal sequential relationship of these types in the sound of this individual was described.     

人工耳蜗的环境声识别研究

对14位正常听力者开展了环境声的人工耳蜗仿真声识别实验,比较了两类声码器仿真(正弦载波和噪声载波)条件下的环境声识别效果差异,然后对9位讲普通话的成年人工耳蜗植入者开展了环境声识别实验。实验材料是从互联网上搜集,并经过12位正常听力者主观测试验证后,筛选出的67种环境声。结果显示,载波类型没有对67种环境声的平均识别效果产生显著影响,但是声学特征的差异会导致单个环境声的识别效果对载波类型有依赖。另外,人工耳蜗植入者的环境声识别效果较差,有待通过信号处理策略、神经接口和康复手段的改进而得到提高。本研究中开发的环境声材料可以用于评估人工耳蜗环境声识别效果。      

Automatic bird sound detection in long real-field recordings: Applications and tools

The primary purpose for pursuing this research is to present a modular approach that enables reliable automatic bird species identification on the basis of their sound emissions in the field. A practical and complete computer-based framework is proposed to detect and time-stamp particular bird species in continuous real field recordings. Acoustic detection of avian sounds can be used for the automatized monitoring of multiple bird taxa and querying in long-term recordings for species of interest for researchers, conservation practitioners, and decision makers, such as environmental indicator taxa and threatened species. This work describes two novel procedures and offers an open modular framework that detects and time-stamps online calls and songs of target bird species and is fast enough to report results in reasonable time for non-processed field recordings of many thousands files and is generic enough to accommodate any species. The framework is evaluated on two large corpora of real field data, targeting the calls and songs of American Robin Turdus migratorius, a Northamerican oscine passerine (true songbird) and the Common Kingfisher Alcedo atthis, a non-passerine species with a wide distribution throughout Eurasia and North Africa. With the aim of promoting the widespread use of digital autonomous recording units (ARUs) and species recognition technologies the processing code and a large corpus of audio recordings is provided in order to enable other researchers to perform and assess comparative experiments.