Acoustic Classes Research Articles

Voice Conversion Using Dynamic Frequency Warping With Amplitude Scaling, for Parallel or Nonparallel Corpora

In Voice Conversion (VC), the speech of a source speaker is modified to resemble that of a particular target speaker. Currently, standard VC approaches use Gaussian mixture model (GMM)-based transformations that do not generate high-quality converted speech due to “over-smoothing” resulting from weak links between individual source and target frame parameters. Dynamic Frequency Warping (DFW) offers an appealing alternative to GMM-based methods, as more spectral details are maintained in transformation; however, the speaker timbre is less successfully converted because spectral power is not adjusted explicitly. Previous work combines separate GMM- and DFW-transformed spectral envelopes for each frame. This paper proposes a more effective DFW-based approach that (1) does not rely on the baseline GMM methods, and (2) functions on the acoustic class level. To adjust spectral power, an amplitude scaling function is used that compares the average target and warped source log spectra for each acoustic class. The proposed DFW with Amplitude scaling (DFWA) outperforms standard GMM and hybrid GMM-DFW methods for VC in terms of both speech quality and timbre conversion, as is confirmed in extensive objective and subjective testing. Furthermore, by not requiring time-alignment of source and target speech, DFWA is able to perform equally well using parallel or nonparallel corpora, as is demonstrated explicitly.

Bed diversity in the shallow water environment of Pappas lagoon in Greece

Acoustic classification systems and the Sediment Trend Analysis method were used to identify and map the bed diversity in a very shallow (<3.0m), coastal lagoon, Pappas lagoon, Western Greece. Analogue acoustic data, collected by means of a 100 kHz side scan sonar system, were digitized and classified into six acoustic classes using recently developed acoustic classification systems (SonarClass, TargAn). By comparing the acoustic classes to ground truth data consisting of sediment grain size and visual inspection of the lagoon-bed, it is demonstrated that the six acoustic classes correlate well with the predominant surface sediment types and vegetation. Thus the spatial distribution of the classes can be considered to represent the spatial pattern of the sedimentary assemblages of the lagoon. The grain size trend analysis identified three dominant sediment pathways and directional trends which could be related to the predominant wind direction, the sediment influx through the inlets and the sediment supply from a small stream in the southern part of the lagoon. The integration of acoustic and sedimentological data together with advanced data processing systems leads to a better understanding of the sedimentary, morphological and biological processes in a shallow lagoon in different spatial and temporal scales and will therefore be beneficial to both sedimentological and biotic-diversity studies.

Marine substratum map of the Causeway Coast, Northern Ireland

A 1:30,000 substratum map for an area off the north coast of Ireland is presented. The study area is bounded in the south by the Causeway coastline and in the north by the following coordinates: top left corner (6°43′36″W, 55°17′N) and top right corner (6°27′W, 55°17′N). This mapping has been made possible through the availability of full seafloor coverage multibeam swath bathymetry and backscatter data (both gridded to 1 m), together with ground-truthing data collected over the past 40 years. Bathymetry data were used to generate terrain indices such as slope, rugosity, aspect, fine- and broad-scale Benthic Position Index, whilst the backscatter data were interpreted visually, subjected to an unsupervised classification process using QTC Multiview, and combined with the bathymetry-derived parameters into a clustermap in ArcGIS. The resulting maps allowed us to divide the seabed into 10 distinct acoustic classes, which, linked to sediment samples, diver surveys, underwater video-tows and remotely operated vehicle surveys, were converted into a substratum map. This is the most accurate seafloor substratum map to date for the north coast of Ireland and could form the basis for more in-depth geological, biological and hydrodynamic studies of this highly dynamic coastline.

Combining echo sounding, sediment and biological data to map benthic habitats in a coastal area of the Persian Gulf

AbstractMapping surveys of coastal benthic habitat in Qeshm Island Geopark, Persian Gulf, were conducted using a combination of biological, sedimentological and echo‐sounding data. The survey area covered approximately 233 km2 in a depth range of 5–25 m, and the data were acquired from a single beam echo sounder, grab, video and still photography. Sediment and macrofauna samples were collected by grab at 76 stations and subjected to classification and ordination analyses. Two acoustic classes were identified differentiating along the near/offshore axis. Sediment texture was dominated by fine grain sizes, with five distinct sub‐sediment types. In total, 214 macrobenthic taxa were identified, of which polychaetes accounted for 60%. Other dominant groups included young sponges, nematodes, malacostracan crustacean, bivalves, ostracods and ophiuroids. Underwater videos and still photos integrated the macrofaunal and sedimentary data and revealed a range of biogenic sedimentary features such as burrows and tubes. The biological data identified six main biological assemblages showing an inshore/offshore pattern. The macrobenthic abundance did not demonstrate a significant difference with depth, although polychaetes were positively correlated with depth. The highest abundance and species richness were observed at median depths. Species distribution and diversity did not show any correlation with sediment type. A preliminary habitat mapping of the south coast of the Qeshm Island Geopark has been carried out, integrating acoustic, sediment and biological data.

Audio segmentation of broadcast news in the Albayzin-2010 evaluation: overview, results, and discussion

Recently, audio segmentation has attracted research interest because of its usefulness in several applications like audio indexing and retrieval, subtitling, monitoring of acoustic scenes, etc. Moreover, a previous audio segmentation stage may be useful to improve the robustness of speech technologies like automatic speech recognition and speaker diarization. In this article, we present the evaluation of broadcast news audio segmentation systems carried out in the context of the Albayzin-2010 evaluation campaign. That evaluation consisted of segmenting audio from the 3/24 Catalan TV channel into five acoustic classes: music, speech, speech over music, speech over noise, and the other. The evaluation results displayed the difficulty of this segmentation task. In this article, after presenting the database and metric, as well as the feature extraction methods and segmentation techniques used by the submitted systems, the experimental results are analyzed and compared, with the aim of gaining an insight into the proposed solutions, and looking for directions which are promising.

Unsupervised fuzzy classification and object-based image analysis of multibeam data to map deep water substrates, Cook Strait, New Zealand

A comprehensive 32 kHz multibeam bathymetry and backscatter survey of Cook Strait, New Zealand (∼8500 km 2), is used to generate a regional substrate classification map over a wide range of water depths, seafloor substrates and geological landforms using an automated mapping method based on the textural image analysis of backscatter data. Full processing of the backscatter is required in order to obtain an image with a strongly attenuated specular reflection. Image segmentation of the merged backscatter and bathymetry layers is constrained using shape, compactness, and texture measures. The number of classes and their spatial distribution are statistically identified by employing an unsupervised fuzzy- c-means (FCM) clustering algorithm to sediment samples, independent of the backscatter data. Classification is achieved from the overlay of the FCM result onto a segmented image and attributing segments with the FCM class. Four classes are identified and uncertainty in class attribution is quantified by a confusion index layer. Validation of the classification map is done by comparing the results with the sediment and structural maps. Backscatter (BS) strength angular profiles are used to show acoustic class separation. The method takes us one step further in combining multibeam data with physical seabed data in a complementary analysis to seek correlations between datasets using object-based image analysis and unsupervised classification. Texture within these identified classes is then examined for correlation with typical backscatter angular responses for mud, sand and gravel. The results show a first order correlation between each of the classes and both the sedimentary properties and the geomorphological map.

Benthic habitat mapping: Concerns using a combined approach (acoustic, sediment and biological data)

Near-shore benthic biotopes, in the 2–30 m depth range, were analyzed and mapped using a combination of biological, sedimentary and acoustic data to interrogate the utility of these methods. The surveyed area covered approximately 270 km 2, along 80 km of the Southeast coast of Portugal. The acoustic data were acquired with a single beam ground discrimination system (QTC VIEW, Series V), connected to a dual frequency echo sounder (50 and 200 kHz). Sediment grain size and macrofaunal community data were obtained by grab sampling at 88 sites. The sedimentary and the biological data were subjected to classification and ordination analysis and the acoustic data obtained with the two frequencies were analyzed individually with the QTC IMPACT (v3.4) software and classified into acoustic classes. The affinity groups obtained with the three datasets were mapped using a Geographic Information System. The study area showed an inshore-offshore pattern with sands ranging from very fine to very coarse, revealing a sediment distribution in the near-shore shelf comprised by two main areas; a very fine/fine sand area up to 7–10 m depth and a very coarse/coarse sand area offshore. In turn, the biological data also showed an inshore-offshore pattern, identifying two main biological assemblages: fine/very fine sand and coarse/very coarse sand communities. Furthermore, the biological pattern clearly reflected the hydrodynamic conditions of the area, from highly hydrodynamic shallow depth to the calmer deeper areas. Accordingly, species richness and densities were lower at shallow areas than at deeper depth. The acoustic data identified the prevailing biosedimentary gradients along the inshore-offshore direction. Overall, the results obtained here showed that the acoustic system could identify two main areas based on sediment grain size, which closely related to the two main biological communities (groups A and B). This study indicates the importance of combining several layers of information in order to increase the spatial resolution of the main biotope distribution (validating the acoustic data) and detail their sub-divisions (ground-truth sampling). Furthermore, we show that designing ground-truth samples on the basis of the acoustic diversity data allows the biological data to “speak” for itself. In general, acoustic techniques are used as a means to optimize ground-truth sampling. This will only be valid if acoustics can capture all the essence of biotope heterogeneity. If this is not the case, special attention should be given when using broad scale methods (acoustic remote sensing) devoted to biotope mapping.

Text Independent Speaker Identification with Finite Multivariate Generalized Gaussian Mixture Model and Hierarchical Clustering Algorithm

In this paper we propose a Text Independent Speaker Identification with Finite Multivariate Generalized Gaussian Mixture Model with Hierarchical Clustering. Each speaker speech spectra are characterized with a mixture of Generalized Gaussian Distribution includes Gaussian and Laplacian distribution as a particular case. It also includes several of the platy, lepto and meso kurtic shapes of the speech spectra. The speech analysis is done with Mel Frequency Cepstral Coefficients extracted from front end process. Using the EM algorithm the model parameters are estimated. The numbers of acoustic classes associated with each speech spectra are determined through Hierarchical clustering. The performance of the proposed algorithm is studied through experimental evolution with 100 speaker‟s data base and found that this algorithm outperforms the existing speaker identification algorithm with GMM. It is also observed that this algorithm performs efficiently even heterogeneous population with small (less than 2 seconds utterances)

Estuarine Habitat Assessment for Construction of a Submarine Transmission Line

The present paper describes a submarine survey using the acoustic discrimination system QTC VIEW (Series V) as an exploratory tool to adjust final route alignment of a new pipeline. By using acoustic sound survey as an exploratory tool described in this paper to adjust final route alignment of a new pipeline to minimize the environmental impact caused and ultimately to avoid any mitigation measures. The transmission pipeline extended from the shore line of Abu-Qir Bay, on the Mediterranean Sea in Egypt, out to 70 nautical miles at sea (60 m water depth). Four main surface sediment types were defined in the study area, namely fine sand, silty sand, silt and clay. Results of the acoustic classification revealed four acoustic classes. The first acoustic class corresponded to fine sand, absence of shell debris and very poor habitats characteristics. The second acoustic class is predominant in the study area and corresponds to the region occupied by silt. It is also characterized by intermediate diversity of macrobenthic invertebrate community which is mainly characterized by polychaeta. The third acoustic class is characterized by silt to silty clay. It is characterized by a high diversity of macrobenthic invertebrate community which is mainly polychaeta with an intermediate diversity of gastropoda and bivalvia. The final acoustic class is characterized by clay and high occurrence of shell debris of gastropoda, bivalvia and polychaeta.

Riverbed sediment classification using multi-beam echo-sounder backscatter data

A method has recently been developed that employs multi-beam echo-sounder backscatter data to both obtain the number of sediment classes and discriminate between them by applying the Bayes decision rule to multiple hypotheses [Simons and Snellen, Appl. Acoust. 70, 1258-1268 (2009)]. In deep water, the number of scatter pixels within the beam footprint is large enough to ensure Gaussian distributions for the backscatter strengths and to increase the discriminative power between acoustic classes. In very shallow water (<10 m), however, this number is too small. This paper presents an extension of this high-frequency methodology for these environments, together with a demonstration of its performance using backscatter data from the river Waal, The Netherlands. The objective of this work is threefold. (i) Increasing the discriminating power of the classification method: high-resolution bathymetry data allow precise bottom slope corrections for obtaining the true incident angle, and the high-resolution backscatter data reduce the statistical fluctuations via an averaging procedure. (ii) Performing a correlation analysis: the dependence of acoustic backscatter classification on sediment physical properties is verified by observing a significant correlation of 0.75 (and a disattenuated correlation of 0.90) between the classification results and sediment mean grain size. (iii) Enhancing the statistical description of the backscatter intensities: angular evolution of the K-distribution shape parameter indicates that the riverbed is a rough surface, in agreement with the results of the core analysis.

Tabletop demonstration of multiple source interference using ultrasonic transducers

Superposition of multiple sources, constructive and destructive interference, and array beam forming are common topics demonstrated in undergraduate acoustic classes. The student learning experience is enhanced by building a system with speakers and measuring the resulting amplitude at various angles with a microphone. This experiment is often performed in an anechoic chamber to achieve quantitative agreement between theory and experiment. We have developed a simple tabletop apparatus using inexpensive ultrasonic transducers that yields comparable results without the need to use an anechoic chamber.

Automated acoustic seabed classification of multibeam images of Stanton Banks

Dividing sidescan images into regions that have similar seabeds is often done by expert interpretation. Automated classification systems are becoming more widely used. This paper describes techniques, based on image amplitudes and texture, that lead to useful and practical automated segmentation of multibeam images. Seabed (or riverbed or lakebed) type affects amplitudes and texture, but so do system operating details and survey geometry. Effects of the last two must be compensated to isolate the effects of seabed type. Images from multibeam surveys are accompanied by bathymetric data from which grazing angles of all sonar footprints can be calculated. By compiling tables of amplitude against range and grazing angle, systematic changes in amplitude with these two variables can be removed consistently. Classification, based on a large number of features, is done in image space to avoid artifacts common in mosaics. Unsupervised segmentation requires clustering, in which records are divided into their natural classes. An objective clustering method using simulated annealing assigns points to classes based on their Bayesian distances from cluster centres. Stanton Banks is a rocky area 100 km north of County Donegal, Ireland, that rises about 100 m above the ocean floor at 180 m. Multibeam images and data from an 80-km 2 survey were classified into regions of acoustic similarity. Assigning labels of physical properties to these regions requires non-acoustic ground truth, which was obtained from a series of 105 photographs. Photographic geological assignments were found to correlate well with the acoustic classes.

Developments in the application of multibeam sonar backscatter for seafloor habitat mapping

Human impacts on the seafloor environment have reached unprecedented levels. To facilitate ocean management and mitigate these impacts, there is a need to improve our understanding of seabed habitats. Recent developments in acoustic survey techniques, in particular multibeam echosounders (MBES), have revolutionised the way we are able to image, map and understand benthic ecosystems. Using MBES, it is now cost-effective to image large areas of the seafloor, and such surveys provide baseline data from which thematic maps of the seabed environment, including maps of benthic habitat, can be derived and interpreted in conjunction with in situ ground-truthing data. This paper provides an overview of recent developments in the application of MBES for seafloor habitat studies, with a focus on the use of backscatter data for surficial geology and habitat mapping. In March 2006, a MBES backscatter workshop brought together a number of international research teams/groups working on novel methods for interpretive/classification routines for segmentation of the backscatter into acoustic classes with the potential to facilitate the delineation of seabed geological and habitat characteristics. This paper introduces the common data set used as part of the workshop, sets out the research context in which the different studies were conducted, and outlines the main themes of the papers presented in Section 6.

Single-beam acoustic ground discrimination of shallow water habitats: 50 kHz or 200 kHz frequency survey?

The single-beam acoustic ground discrimination system QTC View, Series V, was used in the Bay of Cadiz, Southwest Spain, for the identification and mapping of the bottom acoustic diversity. The acoustic data were obtained through two successive surveys, each conducted with one of the following echo sounder frequencies: 50 kHz and 200 kHz. The performance of each survey frequency for the identification of the sedimentary gradients was analyzed. The surveys were conducted during high tide given that the majority of the surveyed area is shallower than 5 m, although depth may occasionally reach 20 m in specific areas located in a navigation channel. The acoustic data obtained at the two different frequencies were, individually, submitted to manual clustering and a final solution consisting of three acoustic classes was reached for both datasets. However, only the geographical distribution of the acoustic classes obtained with 50 kHz echo sounder frequency was coincident with the spatial distribution of the superficial sediment groups (silty medium sand, very silty fine sand and mud), identified through multivariate analysis of the grain-size data of ground-truth sediment samples. The results obtained with the 200 kHz echo sounder frequency did not match the sedimentary gradients obtained for the area surveyed, not even the separation of muddy and sandy areas.

Automatic Speaker Clustering Using a Voice Characteristic Reference Space and Maximum Purity Estimation

This paper investigates the problem of automatically grouping unknown speech utterances based on their associated speakers. In attempts to determine which utterances should be grouped together, it is necessary to measure the voice similarities between utterances. Since most existing methods measure the inter-utterance similarities based directly on the spectrum-based features, the resulting clusters may not be well-related to speakers, but to various acoustic classes instead. This study remedies this shortcoming by projecting utterances onto a reference space trained to cover the generic voice characteristics underlying the whole utterance collection. The resultant projection vectors naturally reflect the relationships of voice similarities among all the utterances, and hence are more robust against interference from nonspeaker factors. Then, a clustering method based on maximum purity estimation is proposed, with the aim of maximizing the similarities between utterances within all the clusters. This method employs a genetic algorithm to determine the cluster to which each utterance should be assigned, which overcomes the limitation of conventional hierarchical clustering that the final result can only reach the local optimum. In addition, the proposed clustering method adapts a Bayesian information criterion to determine how many clusters should be created

Acoustic bottom detection and seabed classification in the German Bight, southern North Sea

To investigate the hydrodynamic activity of the seabed in the German Bight, underwater remote sensing was carried out over an area of 32 km2 located 20 km northeast of Helgoland island in the southern North Sea in January, May and August 2001. On the basis of acoustic seabed classification, six seabed types have been identified by the combined evaluation of side-scan sonar records, wave-shape analysis of echo-sounder data, and 100 grab samples. In five seabed types, the acoustic classes can be distinguished on the basis of sediment characteristics, comprising size components ranging from coarse pebbles to fine sand. The sixth seabed type corresponds to large pebbles and cobbles which are completely overgrown with brown algae. Statistically, the complex spatial patchiness of the six classes varied significantly in the course of the study period. During the winter period (January 2001), the study site was dominated by coarse material, except for a small area of finer sediment in the centre. With the onset of more moderate weather conditions in spring (May 2001), a general fining trend in sediment composition was observed, especially in the deeper western parts of the study area. In summer (August 2001), finer sediments still dominated but a slight increase in signal roughness suggests an overprint by coarser lag deposits and/or denser coverage by benthic organisms (e.g. Lanice conchilega) which then were found more frequently in grab samples, in association with finer sand. These findings demonstrate that the distribution of seafloor sediments and their benthic fauna in the deeper part of the German Bight region are controlled largely by seasonal changes in hydrodynamic conditions. These changes are reflected in correspondingly high variability in the complex patchiness of sediment distribution patterns, which would not have been adequately resolved by any standard sampling procedure.

Sea-bottom classification across a shallow-water bar channel and near-shore shelf, using single-beam acoustics

An acoustic ground discrimination system (QTC VIEW, Series IV) was used to identify and map the bottom acoustic diversity in the bar channel of Ria de Aveiro, Western Portugal. The majority of the survey area presented shallow depth for this type of equipment, ranging mainly from 5 to 15 m. Depth occasionally reached 25 m in specific areas located across the entrance channel, dug by the strong tidal currents, reaching 3 m/s. The acoustic data were submitted to manual and auto-cluster and the results obtained from both procedures were coherent. Using aids to the acoustic classification and ground-truth sediment data, a final solution consisting of four acoustic classes was reached. Their geographical distribution was coincident with the spatial distribution of the major bottom types and sediment groups (hard bottom, coarse sand, medium sand and fine sand), identified through multivariate analysis of the grain-size data, and reflected the complex hydrodynamics of the entrance channel. The acoustic pattern was coincident at the intersections of the acoustic survey lines, assuring the repeatability of the acoustic procedure. Overall, the acoustic approach showed consistent results for the assessment and mapping of the benthic habitats in this shallow-water coastal area, providing a very valuable tool in an area where conventional sediment sampling is less favourable, namely due to strong tidal currents and frequent ship traffic, such as the entrance channel of Ria de Aveiro and the near-shore adjacent shelf.

Acoustic seabed classification in a coastal environment (outer Weser Estuary, German Bight)—a new approach to monitor dredging and dredge spoil disposal

Acoustic seabed classification provides a sophisticated technique to discriminate seabed characteristics and to map their distribution at high spatial and temporal resolution. In the present study, the seabed classification system QTC View/Impact was applied to investigate a 9- km 2 area at water depths between 6 and 20 m in the outer Weser Estuary (German Bight, southeastern North Sea). The survey area comprises parts of a routinely dredged shipping channel as well as of a dredge spoil disposal site. The acoustic data, collected by means of a singlebeam 200-kHz echosounder combined with a QTC system, were classified into three acoustic classes. These classes are identified with (1) fine to medium sand, with a low content of shell fragments; (2) medium sand, with a moderate content of shell fragments; and (3) medium to coarse sand, with a high content of shell fragments. By comparing this point-related acoustic classification with sidescan sonographs, it can be convincingly demonstrated that the acoustic data obtained by the QTC system reflect more than simply sediment type distribution, but that rather the occurrence of bedforms also appears to be important. Furthermore, in a time series of three QTC surveys in a period of 14 months, it is apparent that dredging as well as the disposal of dredge spoil constitute a strong impact on the sediment distribution and seabed morphology of the study site.

Acoustic detection of a scallop bed from a single-beam echosounder in the St. Lawrence

Abstract Single-beam seabed echoes combined with epi-macrobenthos photographs were used to remotely detect a scallop bed and characterize the specific acoustic signal of Iceland scallop (Chlamys islandica). A dense scallop bed was surveyed in 2002, with a QTC VIEW Series IV acoustic ground-discrimination system (AGDS) connected to a 38 kHz, 7° split-beam SIMRAD EK60 scientific echosounder. In 2003, a 50 kHz, 42° single-beam SUZUKI ES-2025 echosounder was connected to a QTC VIEW Series V AGDS. The QTC VIEW data were analysed with QTC IMPACT following the standard procedures and classified into acoustic classes. Several approaches were tested: unsupervised and supervised survey strategies directed to specific benthic communities. The SIMRAD EK60 seabed volume-backscattering strength (Sv) was submitted to a principal component analysis (PCA), before and after removal of a depth trend, and the scores on the first 10 principal components were classed by a K-means cluster analysis. The same seabed Sv data were submitted to stepwise discriminant analysis whose training data sets were defined with the ground-truth photographs using different groupings: biotope types, community types, and finally scallop-density classes. All the QTC AGDS approaches failed to reveal the scallop bed, community structures, or biotopes. The QTC classifications mimicked the bathymetry with a strong correlation of the acoustic classes with depth. The seabed Sv PCA + K-means approach presented similar depth-dependence, but, the PCA + K-means on the Sv residuals revealed the scallop bed. The discriminant analysis was the best solution for the scallop density with a general classification success rate of 75% and up to 91% for the highest density class. The Sv signature of the scallop bed is presented, and the most discriminant part of the acoustic signal is identified.

Acoustic characterization of developmental speech disorders

A novel approach to classifying children with developmental speech delays (DSD) involving /r/ was developed. The approach first derives an acoustic classification of /r/ tokens based on their forced Viterbi alignment to a five-state hidden Markov model (HMM) of normally articulated /r/. Children with DSD are then classified in terms of the proportion of their /r/ productions that fall into each broad acoustic class. This approach was evaluated using 953 examples of /r/ as produced by 18 DSD children and an approximately equal number of /r/ tokens produced by a much larger number of normally articulating children. The acoustic classification identified three broad categories of /r/ that differed substantially in how they aligned to the normal speech /r/ HMM. Additionally, these categories tended to partition tokens uttered by DSD children from those uttered by normally articulating children. Similarities among the DSD children and average normal child measured in terms of the proportion of their /r/ productions that fell into each of the three broad acoustic categories were used to perform a hierarchical clustering. This clustering revealed groupings of DSD children who tended to approach /r/ production in one of several acoustically distinct manners.