BACKGROUND: Current high-throughput sequencing platforms provide capacity to sequence multiple samples in parallel. Different samples are labeled by attaching a short sample specific nucleotide sequence, barcode, to each DNA molecule prior pooling them into a mix containing a number of libraries to be sequenced simultaneously. After sequencing, the samples are binned by identifying the barcode sequence within each sequence read. In order to tolerate sequencing errors, barcodes should be sufficiently apart from each other in sequence space. An additional constraint due to both nucleotide usage and basecalling accuracy is that the proportion of different nucleotides should be in balance in each barcode position. The number of samples to be mixed in each sequencing run may vary and this introduces a problem how to select the best subset of available barcodes at sequencing core facility for each sequencing run. There are plenty of tools available for de novo barcode design, but they are not suitable for subset selection. RESULTS: We have developed a tool which can be used for three different tasks: 1) selecting an optimal barcode set from a larger set of candidates, 2) checking the compatibility of user-defined set of barcodes, e.g. whether two or more libraries with existing barcodes can be combined in a single sequencing pool, and 3) augmenting an existing set of barcodes. In our approach the selection process is formulated as a minimization problem. We define the cost function and a set of constraints and use integer programming to solve the resulting combinatorial problem. Based on the desired number of barcodes to be selected and the set of candidate sequences given by user, the necessary constraints are automatically generated and the optimal solution can be found. The method is implemented in C programming language and web interface is available at http://ekhidna2.biocenter.helsinki.fi/barcosel . CONCLUSIONS: Increasing capacity of sequencing platforms raises the challenge of mixing barcodes. Our method allows the user to select a given number of barcodes among the larger existing barcode set so that both sequencing errors are tolerated and the nucleotide balance is optimized. The tool is easy to access via web browser.

MOTIVATION: When targeted to a barcoding region, high-throughput sequencing can be used to identify species or operational taxonomical units from environmental samples, and thus to study the diversity and structure of species communities. Although there are many methods which provide confidence scores for assigning taxonomic affiliations, it is not straightforward to translate these values to unbiased probabilities. We present a probabilistic method for taxonomical classification (PROTAX) of DNA sequences. Given a pre-defined taxonomical tree structure that is partially populated by reference sequences, PROTAX decomposes the probability of one to the set of all possible outcomes. PROTAX accounts for species that are present in the taxonomy but that do not have reference sequences, the possibility of unknown taxonomical units, as well as mislabeled reference sequences. PROTAX is based on a statistical multinomial regression model, and it can utilize any kind of sequence similarity measures or the outputs of other classifiers as predictors. RESULTS: We demonstrate the performance of PROTAX by using as predictors the output from BLAST, the phylogenetic classification software TIPP, and the RDP classifier. We show that PROTAX improves the predictions of the baseline implementations of TIPP and RDP classifiers, and that it is able to combine complementary information provided by BLAST and TIPP, resulting in accurate and unbiased classifications even with very challenging cases such as 50% mislabeling of reference sequences. AVAILABILITY AND IMPLEMENTATION: Perl/R implementation of PROTAX is available at http://www.helsinki.fi/science/metapop/Software.htm CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

This paper is related to the development of signal processing techniques for automatic recognition of bird species. Three different parametric representations are compared. The first representation is based on sinusoidal modeling which has been earlier found useful for highly tonal bird sounds. Mel-cepstrum parameters are used since they have been found very useful in the parallel problem of speech recognition. Finally, a vector of various descriptive features is tested because such models are popular in audio classification applications, and bird song is almost like music. We briefly introduce the methods and evaluate their performance in the classification and recognition of both individual syllables and song fragments of 14 common North-European Passerine bird species

The article is related to the development of techniques for automatic recognition of bird species by their sounds. It has been demonstrated earlier that a simple model of one time-varying sinusoid is very useful in classification and recognition of typical bird sounds. However, a large class of bird sounds are not pure sinusoids but have a clear harmonic spectrum structure. We introduce a way to classify bird syllables into four classes by their harmonic structure.

Bird song can be divided into a sequence of syllabic elements. We investigate the possibility of bird species recognition based on the syllable pair histogram of the song. This representation compresses the variable-length syllable sequence into a fixed-dimensional feature vector. The histogram is computed by means of Gaussian syllable prototypes which are automatically found given the song data and the dissimilarity measure of syllables. Our representation captures the use of the syllable alphabet and also some temporal structure of the song. We demonstrate the method in bird species recognition with song patterns obtained from fifty individuals belonging to four common passerine bird species.

The present work compares three different methods for training acoustic models in a Finnish large vocabulary speech recognition system. The models are trained using the maximum likelihood (ML), maximum a posteriori (MAP), and variational Bayesian (VB) principle. The results show that when the model complexity is properly chosen, all three methods give similar performance. As the model complexity increases, the perfor-mance of ML based system starts to degrade whereas no over-fitting is observed using MAP and VB based models. MAP gives slightly better recognition accuracy over VB but it cannot be used for model selection without auxiliary data. The advan-tage of VB is that it can be used for selecting a well performing model structure using only training data. 1.

Feature extraction is the key element when aiming at robust speech recognition. Both linear and nonlinear data-driven feature transformations are applied to the logarithmic mel-spectral context feature vectors in the TIMIT phone recognition task. Transformations are based on principal component analysis (PCA), independent component analysis (ICA), linear discriminant analysis (LDA) and multilayer perceptron network based nonlinear discriminant analysis (NLDA). All four methods outperform the baseline system which consists of the standard feature representation based on MFCCs (mel-frequency cepstral coefficients) with the first-order deltas, using a mixture-of-Gaussians HMM recognizer. Further improvement is gained by forming the feature vector as a concatenation of the outputs of all four feature transformations.

In this work, linear and nonlinear feature transformations have been experimented in ASR front end. Unsupervised transformations were based on principal component analysis and independent component analysis. Discriminative transformations were based on linear discriminant analysis and multilayer perceptron networks. The acoustic models were trained using a subset of HUB5 training data and they were tested using OGI Numbers corpus. Baseline feature vector consisted of PLP cepstrum and energy with first and second order deltas. None of the feature transformations could outperform the baseline when used alone, but improvement in the word error rate was gained when the baseline feature was combined with the feature transformation stream. Two combination methods were experimented: feature vector concatenation and n-best list combination using ROVER. Best results were obtained using the combination of the baseline PLP cepstrum and the feature transform based on multilayer perceptron network. The word error rate in the number recognition task was reduced from 4.1 to 3.1.

A method is presented for speeding up the performance of the HMM based speech recognition system where the states are modeled by a large number of Gaussian kernels. The emission probabilities of the states are usually dominated by the nearest Gaussians to the input vector. The speedup is gained without deteriorating the recognition accuracy by concentrating on these kernels in the reduced K-best-kernel search. In this work, the time information of the input is encoded to the connections of the kernels. The search for the dominating kernels is then performed along the kernel connections which model the trajectories of the speech in the feature space. In the experiments, speaker-dependent speech recognizers were trained for ten speakers. The number of distance computations between feature vectors and kernel mean vectors was reduced 75% without increasing the average phoneme recognition error, which was 5.7% for the baseline system.

This paper presents an unsupervised segmentation method for feature sequences based on competitive-learning hidden Markov models. Models associated with the nodes of the self-organizing map learn to become selective to the segments of temporal input sequences. Input sequences may have arbitrary lengths. Segment models emerge then on the map through an unsupervised learning process. The method was tested in speech recognition, where the performance of the emergent segment models was as good as the performance of the traditionally used linguistic speech segment models. The benefits of the proposed method are the use of unsupervised learning for obtaining the state models for temporal data and the convenient visualization of the state space on the two-dimensional map.

Time information of the input data is used for evaluating the goodness of the self-organizing map to store and represent temporal feature vector sequences. A new node neighborhood is defined for the map which takes the temporal order of the input samples into account. A connection is created between those two map modes which are the best-matching units for two successive input samples in time. This results in the time-topology preserving network.

It has been hypothesized that the Finnish language is well suited to speech-to-text conversion for the communication aids of the hearing impaired. In a related study it was shown that, depending on context, 10 to 20 % of phoneme errors can be tolerated with good comprehension when reading text converted from raw phonemic recognition. Two sets of phoneme recognition experiments were carried out in this study in order to evaluate the performance of existing speech recognition systems in this application. For telephone bandwidth speech both systems showed speaker dependent error scores of about 10 % or below, thus supporting the feasibility of the application. For speaker independent cases the error rate was typically more than 20 % which is too high for effortless and fluent communication. 1.