Automated categorization of bioacoustic signals: avoiding perceptual pitfalls

Dividing the acoustic repertoires of animals into biologically relevant categories presents a widespread problem in the study of animal sound communication, essential to any comparison of repertoires between contexts, individuals, populations, or species. Automated procedures allow rapid, repeatable, and objective categorization, but often perform poorly at detecting biologically meaningful sound classes. Arguably this is because many automated methods fail to address the nonlinearities of animal sound perception. We present a new method of categorization that incorporates dynamic time-warping and an adaptive resonance theory (ART) neural network. This method was tested on 104 randomly chosen whistle contours from four captive bottlenose dolphins (Tursiops truncatus), as well as 50 frequency contours extracted from calls of transient killer whales (Orcinus orca). The dolphin data included known biologically meaningful categories in the form of 42 stereotyped whistles produced when each individual was isolated from its group. The automated procedure correctly grouped all but two stereotyped whistles into separate categories, thus performing as well as human observers. The categorization of killer whale calls largely corresponded to visual and aural categorizations by other researchers. These results suggest that this methodology provides a repeatable and objective means of dividing bioacoustic signals into biologically meaningful categories.     

一种宽吻海豚通讯信号自动分类的方法

针对宽吻海豚通讯信号自动分类提出了一种基于句法模式识别的方法。该方法首先提取海豚通讯信号基频随时间变化的轨迹曲线,然后提取基频变化的基元序列。根据海豚通讯信号分类的标准,归纳出产生各类海豚通讯信号基频基元序列的文法。对未知类别的海豚通讯信号,提取其基频变化的基元序列,根据各类模式的文法对基元序列进行分类,进而实现海豚通讯信号的自动分类。实验结果显示本文方法的分类准确率达到了95%。本文方法预期为海豚生物学行为的声学研究提供一定的技术支持。      

Characteristics of whistles from the acoustic repertoire of resident killer whales (Orcinus orca) off Vancouver Island, British Columbia

The acoustic repertoire of killer whales (Orcinus orca) consists of pulsed calls and tonal sounds, called whistles. Although previous studies gave information on whistle parameters, no study has presented a detailed quantitative characterization of whistles from wild killer whales. Thus an interpretation of possible functions of whistles in killer whale underwater communication has been impossible so far. In this study acoustic parameters of whistles from groups of individually known killer whales were measured. Observations in the field indicate that whistles are close-range signals. The majority of whistles (90%) were tones with several harmonics with the main energy concentrated in the fundamental. The remainder were tones with enhanced second or higher harmonics and tones without harmonics. Whistles had an average bandwidth of 4.5 kHz, an average dominant frequency of 8.3 kHz, and an average duration of 1.8 s. The number of frequency modulations per whistle ranged between 0 and 71. The study indicates that whistles in wild killer whales serve a different function than whistles of other delphinids. Their structure makes whistles of killer whales suitable to function as close-range motivational sounds.     

A method for classifying whistles of bottlenose dolphin(Tursiops truncates) based on syntactic pattern reorganization

A method based on syntactic pattern recognition was presented to automatically classify whistles of bottlenose dolphin.Dolphin whistles have typically been characterized in terms of their instantaneous frequency as a function of time,which is also known as "whistle contour".The frequency variation features of a whistle were extracted according to its contour.Then,the frequency variation features were used for learning grammatical patterns.A whistle was classified according to grammatical pattern of its frequency variation features.The experimental results showed that the classification accuracy of the proposed method was 95%.The method can provide technical support for acoustic study of dolphins' biological behavior.     

Synthesis and modification of the whistles of the bottlenose dolphin, Tursiops truncatus

A signal-processing algorithm was developed to analyze harmonic frequency-modulated sounds, to modify the parameters of the analyzed signal, and to synthesize a new analytically specified signal that resembles the original signal in specified features. This algorithm was used with dolphin whistles, a frequency-modulated harmonic signal that has typically been described in terms of its contour, or pattern of modulation of the fundamental frequency. In order to test whether other features may also be salient to dolphins, the whistle analysis calculates the energies at the harmonics as well as the fundamental frequency of the whistle. The modification part of the algorithm can set all of these energies to a constant, can shift the whistle frequency, and can expand or compress the time base or the frequency of the whistle. The synthesis part of the algorithm then synthesizes a waveform based upon the energies and frequencies of the fundamental and first two harmonics. These synthetic whistles will be useful for evaluating what acoustic features dolphins use in discriminating different whistles.     

The whistles of Hawaiian spinner dolphins

The characteristics of the whistles of Hawaiian spinner dolphins (Stenella longirostris) are considered by examining concurrently the whistle repertoire (whistle types) and the frequency of occurrence of each whistle type (whistle usage). Whistles were recorded off six islands in the Hawaiian Archipelago. In this study Hawaiian spinner dolphins emitted frequency modulated whistles that often sweep up in frequency (47% of the whistles were upsweeps). The frequency span of the fundamental component was mainly between 2 and 22 kHz (about 94% of the whistles) with an average mid-frequency of 12.9 kHz. The duration of spinner whistles was relatively short, mainly within a span of 0.05 to 1.28 s (about 94% of the whistles) with an average value of 0.49 s. The average maximum frequency of 15.9 kHz obtained by this study is consistent with the body length versus maximum frequency relationship obtained by Wang et al. (1995a) when using spinner dolphin adult body length measurements. When comparing the average values of whistle parameters obtained by this and other studies in the Island of Hawaii, statistically significant differences were found between studies. The reasons for these differences are not obvious. Some possibilities include differences in the upper frequency limit of the recording systems, different spinner groups being recorded, and observer differences in viewing spectrograms. Standardization in recording and analysis procedure is clearly needed.     

The effect of recording and analysis bandwidth on acoustic identification of delphinid species

Because many cetacean species produce characteristic calls that propagate well under water, acoustic techniques can be used to detect and identify them. The ability to identify cetaceans to species using acoustic methods varies and may be affected by recording and analysis bandwidth. To examine the effect of bandwidth on species identification, whistles were recorded from four delphinid species (Delphinus delphis, Stenella attenuata, S. coeruleoalba, and S. longirostris) in the eastern tropical Pacific ocean. Four spectrograms, each with a different upper frequency limit (20, 24, 30, and 40 kHz), were created for each whistle (n = 484). Eight variables (beginning, ending, minimum, and maximum frequency; duration; number of inflection points; number of steps; and presence/absence of harmonics) were measured from the fundamental frequency of each whistle. The whistle repertoires of all four species contained fundamental frequencies extending above 20 kHz. Overall correct classification using discriminant function analysis ranged from 30% for the 20-kHz upper frequency limit data to 37% for the 40-kHz upper frequency limit data. For the four species included in this study, an upper bandwidth limit of at least 24 kHz is required for an accurate representation of fundamental whistle contours.     

Characteristics of whistles from rough-toothed dolphins (Steno bredanensis) in Rio de Janeiro coast, southeastern Brazil

There is no information about the whistles of rough-toothed dolphins in the South Atlantic Ocean. This study characterizes the whistle structure of free-ranging rough-toothed dolphins recorded on the Rio de Janeiro coast, southeastern Brazil, and compares it to that of the same species in other regions. A total of 340 whistles were analyzed. Constant (N = 115; 33.8%) and ascending (N = 99; 29.1%) whistles were the most common contours. The whistles recorded had their fundamental frequencies between 2.24 and 13.94 kHz. Whistles without inflection points were frequently emitted (N = 255; 75%). Some signals presented breaks or steps in their contour (N = 97; 28.5%). Whistle duration was short (347 ± 236 ms and 89.7% of the whistles lasted <600 ms). Seventy-eight whistle contour types were found in the total of whistles analyzed, and 27 (7.9%) of these occurred only once. Most of the whistle types were unique to a particular recording session (N = 43). The signals emitted by the rough-toothed dolphins in southeastern Brazil were characterized by low frequency modulation, short duration, low number of inflection points, and breaks. Differences in the mean values of the whistle parameters were found between this and other studies that recorded Steno bredanensis, but as in other localities, whistles above 14 kHz are rare.     

Estimated communication range of social sounds used by bottlenose dolphins (Tursiops truncatus)

Bottlenose dolphins, Tursiops truncatus, exhibit flexible associations in which the compositions of groups change frequently. We investigated the potential distances over which female dolphins and their dependent calves could remain in acoustic contact. We quantified the propagation of sounds in the frequency range of typical dolphin whistles in shallow water areas and channels of Sarasota Bay, Florida. Our results indicated that detection range was noise limited as opposed to being limited by hearing sensitivity. Sounds were attenuated to a greater extent in areas with seagrass than any other habitat. Estimates of active space of whistles showed that in seagrass shallow water areas, low-frequency whistles (7-13 kHz) with a 165 dB source level could be heard by dolphins at 487 m. In shallow areas with a mud bottom, all whistle frequency components of the same whistle could be heard by dolphins travel up to 2 km. In channels, high-frequency whistles (13-19 kHz) could be detectable potentially over a much longer distance (> 20 km). Our findings indicate that the communication range of social sounds likely exceeds the mean separation distances between females and their calves. Ecological pressures might play an important role in determining the separation distances within communication range.     

Captive dolphins,Tursiops truncatus,develop signature whistles that match acoustic features of human-made model sounds

This paper presents a cross-sectional study testing whether dolphins that are born in aquarium pools where they hear trainers' whistles develop whistles that are less frequency modulated than those of wild dolphins. Ten pairs of captive and wild dolphins were matched for age and sex. Twenty whistles were sampled from each dolphin. Several traditional acoustic features (total duration, duration minus any silent periods, etc.) were measured for each whistle, in addition to newly defined flatness parameters: total flatness ratio (percentage of whistle scored as unmodulated), and contiguous flatness ratio (duration of longest flat segment divided by total duration). The durations of wild dolphin whistles were found to be significantly longer, and the captive dolphins had whistles that were less frequency modulated and more like the trainers' whistles. Using a standard t-test, the captive dolphin had a significantly higher total flatness ratio in 9/10 matched pairs, and in 8/10 pairs the captive dolphin had significantly higher contiguous flatness ratios. These results suggest that captive-born dolphins can incorporate features of artificial acoustic models made by humans into their signature whistles.     

Geographic variations in the whistles of spinner dolphins (Stenella longirostris) of the Main Hawai'ian Islands

Geographic variations in the whistles of Hawai'ian spinner dolphins are discussed by comparing 27 spinner dolphin pods recorded in waters off the Islands of Kaua'i, O'ahu, Lana'i, and Hawai'i. Three different behavioral states, the number of dolphins observed in each pod, and ten parameters extracted from each whistle contour were considered by using clustering and discriminant function analyses. The results suggest that spinner dolphin pods in the Main Hawai'ian Islands share characteristics in approximately 48% of their whistles. Spinner dolphin pods had similar whistle parameters regardless of the island, location, and date when they were sampled and the dolphins' behavioral state and pod size. The term "whistle-specific subgroup" (WSS) was used to designate whistle groups with similar whistles parameters (which could have been produced in part by the same dolphins). The emission rate of whistles was higher when spinner dolphins were socializing than when they were traveling or resting, suggesting that whistles are mainly used during close-range interactions. Spinner dolphins also seem to vary whistle duration according to their general behavioral state. Whistle duration and the number of turns and steps of a whistle may be more important in delivering information at the individual level than whistle frequency parameters.     

Who is whistling? Localizing and identifying phonating dolphins in captivity

Acoustic communication through whistles is well developed in dolphins. However, little is known on how dolphins are using whistles because localizing the sound source is not an easy task. In the present study, the hyperbola method was used to localize the sound source using a two-hydrophone array. A combined visual and acoustic method was used to determine the identity of the whistling dolphin. In an aquarium in Mexico City where two adult bottlenose dolphins were housed we recorded 946 whistles during 22 days. We found that a dolphin was located along the calculated hyperbola for 72.9% of the whistles, but only for 60.3% of the whistles could we determine the identity of the whistling dolphin. However, sometimes it was possible to use other cues to identify the whistling dolphin. It could be the animal that performed a behavior named “observation” at the time whistling occurred or, when a whistle was only recorded on one channel, the whistling dolphin could be the animal located closest to the hydrophone that captured the whistle. Using these cues, 15.4% of the whistles were further ascribed to either dolphin to obtain an overall identification efficiency of 75.7%. Our results show that a very simple and inexpensive acoustic setup can lead to a reasonable number of identifications of the captive whistling dolphin: this is the first study to report such a high rate of whistles identified to the free swimming, captive dolphin that produced them. Therefore, we have a data set with which we can investigate how dolphins are using whistles. This method can be applied in other aquaria where a small number of dolphins is housed; though, the actual efficiency of this method will depend on how often dolphins spend time next to each other and on the reverberation conditions of the pool.     

Source levels and harmonic content of whistles in white-beaked dolphins (Lagenorhynchus albirostris)

Recordings of white-beaked dolphin whistles were made in Faxafl6i Bay (Iceland) using a three-hydrophone towed linear array. Signals from the hydrophones were routed through an amplifier to a lunch box computer on board the boat and digitized using a sample rate of 125 kHz per channel. Using this method more than 5000 whistles were recorded. All recordings were made in sea states 0-1 (Beaufort scale). Dolphins were located in a 2D horizontal plane by using the difference of arrival time to the three hydrophones, and source levels were estimated from these positions using two different methods (I and II). Forty-three whistles gave a reliable location for the vocalizing dolphin when using method II and of these 12 when using method I. Source level estimates on the center hydrophone were higher using method I [average source level 148 (rms) +/- 12 dB, n = 36] than for method II [average source level 139 (rms) +/- 12 dB, n = 36]. Using these rms values the maximum possible communication range for whistling dolphins given the local ambient noise conditions was then estimated. The maximum range was 10.5 km for a dolphin whistle with the highest source level (167 dB) and about 140 m for a whistle with the lowest source level (118 dB). Only two of the 43 whistles contained an unequal number of harmonics recorded at the three hydrophones judging from the spectrograms. Such signals could be used to calculate the directionality of whistles, but more recordings are necessary to describe the directionality of white-beaked dolphin whistles.     

The rodent ultrasound production mechanism.

Rodents produce two types of sounds, audible and ultrasonic, that differ markedly in physical structure. Studies of sound production in light gases show that whereas the audible cries appear to be produced, as in the case of most other mammals, by vibrating structures in the larynx, the ultrasonic cries are produced by a different mechanism, probably a whistle. 'Bird-call' whistles are shown to have all the properties of rodent ultrasonic cries and to mimic them in almost every detail. Thus it is concluded that rodents have two distinct sound production mechanisms, one for audible cries and one for ultrasonic cries.     

基于子频带能量特征提取的汽车鸣笛声识别*

针对城市中汽车违法鸣笛声之间识别分类较难的问题，为了快速准确的识别鸣笛声并将不同种鸣笛声之间进行分类，在鸣笛声识别分类中提出了应用子频带能量提取鸣笛声的特征，并利用BP神经网络对提取的子频带能量特征值矩阵进行学习训练，且在神经网络学习过程中利用可变学习速度的方法，减小了神经网络的迭代次数。实验表明利用此种子频带能量特征提取法使鸣笛声与非鸣笛声的平均识别率达到了94.889%；使不同鸣笛声之间的分类正确率最大达到了93.75%，实现了不同鸣笛声之间的分类。     

The freshwater dolphin Inia geoffrensis geoffrensis produces high frequency whistles

Because whistles are most commonly associated with social delphinids, they have been largely overlooked, ignored, or presumed absent, in solitary freshwater dolphin species. Whistle production in the freshwater dolphin, the boto (Inia geoffrensis geoffrensis), has been controversial. Because of its sympatry with tucuxi dolphins (Sotalia fluviatilis), a whistling species, some presume tucuxi whistles might have been erroneously assigned to the boto. Using a broadband recording system, we recorded over 100 whistles from boto dolphins in the Yasunf River, Ecuador, where the tucuxi dolphins are absent. Our results therefore provide conclusive evidence for whistle production in Inia geoffrensis geoffrensis. Furthermore, boto whistles are significantly different from tucuxi whistles recorded in nearby rivers. The Ecuadorian boto whistle has a significantly greater frequency range (5.30-48.10 kHz) than previously reported in other populations (Peru and Colombia) that were recorded with more bandwidth limited equipment. In addition, the top frequency and the range are greater than in any other toothed whale species recorded to date. Whistle production was higher during resting activities, alone or in the presence of other animals. The confirmation of whistles in the boto has important implications for the evolution of whistles in Cetacea and their association with sociality.     

Experimental demonstration of underwater acoustic communication using bionic signals

This paper applies dolphin whistles to covert underwater acoustic (UWA) communication and proposes a UWA communication scheme based on M-ary bionic signal coding. At the transmitter end, the scheme maps multiple information bits into a dolphin whistle through a signal selector. At the receiver end, passive time reversal mirror (PTRM) is used for channel equalization and source information is restored according to the decision of which whistle is transmitted. The scheme has high spread spectrum gain. The anti multi-path performance is greatly improved when using PTRM. Different from traditional covert UWA communication methods, this mimicked signal is unlikely to alert an adversary even in high SNRs because of its real existence in marine environment. A tank experiment is conducted for the scheme, at communication rate of 50 bit/s with SNR −5 dB user information is recovered at a very low bit error rate. The results of tank experiment demonstrate the feasibility of this covert UWA communication scheme.     

印太瓶鼻海豚(Tursiops aduncus)通讯声信号分类及特征参数的环境差异性分析

通过长期记录室内水池环境下两只印太瓶鼻海豚通讯信号,并与海湾自然环境下同样的两只海豚所发出的通讯信号进行比较分析,从信号类型、声谱特征等方面研究生活环境变化下瓶鼻海豚通讯信号的差异性。结果表明,生活环境的差异,会改变瓶鼻海豚通讯信号。海湾自然环境下,瓶鼻海豚通讯信号以正弦型信号为主;而室内水池环境下,上扫型信号比例明显增多,而正弦型信号减少。两种环境下,瓶鼻海豚通讯信号在持续时间、拐点数、起始频率、结束频率、最小频率、最大频率等存在显著性差异(p<0.05),但信号的频率变化量相近(p=0.29)。结果为提高海豚通讯信号认知和增强海豚生物行为研究提供一定的科学参考,同时也为仿生隐蔽通信提供技术支撑。