Eugene Coyle

Extended Nonnegative Tensor Factorisation Models for Musical Sound Source Separation

Recently, shift-invariant tensor factorisation algorithms have been proposed for the purposes of sound source separation of pitched musical instruments. However, in practice, existing algorithms require the use of log-frequency spectrograms to allow shift invariance in frequency which causes problems when attempting to resynthesise the separated sources. Further, it is difficult to impose harmonicity constraints on the recovered basis functions. This paper proposes a new additive synthesis-based approach which allows the use of linear-frequency spectrograms as well as imposing strict harmonic constraints, resulting in an improved model. Further, these additional constraints allow the addition of a source filter model to the factorisation framework, and an extended model which is capable of separating mixtures of pitched and percussive instruments simultaneously.

Clustering NMF basis functions using Shifted NMF for monaural sound source separation

Non-negative Matrix Factorization (NMF) has found use in single channel separation of audio signals, as it gives a parts-based decomposition of audio spectrograms where the parts typically correspond to individual notes or chords. However, a notable shortcoming of NMF is the need to cluster the basis functions to their sources after decomposition. Despite recent improvements in algorithms for clustering the basis functions to sources, much work still remains to further improve these algorithms. To this end we present a novel clustering algorithm which overcomes some of the limitations of previous clustering methods. This involves the use of Shifted Nonnegative Matrix Factorization (SNMF) as a means of clustering the frequency basis functions obtained from NMF. Results show that this gives improved clustering of pitched basis functions over previous methods.

Shifted non-negative matrix factorisation for sound source separation

A shifted non-negative matrix factorisation algorithm is derived, which offers advantages over previous matrix factorisation techniques for the purposes of single channel source separation. It represents a sound source as translations of a single frequency basis function. These translations approximately correspond to notes played by an instrument. Results are presented for a set of synthetic data, and on a single channel recording of piano and clarinet. Though the system is aimed at musical recordings, the technique can be applied to any data which contains shifted versions of an underlying factor, and so the algorithm could possibly be used in other applications such as image processing

Sound Source Separation Using Shifted Non-Negative Tensor Factorisation

Recently, shifted non-negative matrix factorisation was developed as a means of separating harmonic instruments from single channel mixtures. However, in many cases two or more channels are available, in which case it would be advantageous to have a multichannel version of the algorithm. To this end, a shifted non-negative tensor factorisation algorithm is derived, which extends shifted non-negative matrix factorisation to the multi-channel case. The use of this algorithm for multi-channel sound source separation of harmonic instruments is demonstrated. Further, it is shown that the algorithm can be used to perform non-negative tensor deconvolution, a multi-channel version of non-negative matrix deconvolution, to separate sound sources which have time evolving spectra from multi-channel signals

Speech-adaptive time-scale modification for computer assisted language-learning

In traditional foreign language learning programmes students are offered a tutor model characterised by slow, deliberate speech. This is insufficient to prepare them to cope with native, colloquial speech. By applying a time-scale modification (TSM) algorithm to natural-speed, native speech, students have access to a more desirable, natural speech corpus which permits them to practise essential listening skills in a more focussed manner. We present a method which allows slowing down speech without compromising the quality, pitch or naturalness of the slowed speech by applying different scaling factors to different types of speech segments. The new method is compared to traditional uniform-rate techniques and other variable time-scaling methods. The results show that the proposed approach produces a superior quality output, even for high modification rates.

Interacting with large music collections: Towards the use of environmental metadata

Large music collections afford the listener flexibility in the form of choice, which enables the listener to choose the appropriate piece of music to enhance or complement their listening scenario on-demand. However, bundled with such a large music collection is the demanding task of manually searching through each entry in the collection to find the appropriate song required by the listener. This paper highlights the need for contextual and environmental information, which ultimately defines the listenerpsilas listening scenario. Here, the preliminary results of an online music survey are analysed. These results indicate the possibility of how environmental features may be used as metadata to indicate the listenerpsilas mood. Therefore, environmental features, such as location, activity, temperature, lighting and weather have great potential as metadata and hence may be used to create a personalised automatic playlist generator for large music collections.

Onset detection using comb filters

A technique for detecting note onsets using FIR comb filters which have different filter delays is presented. The proposed onset detector focuses on the inharmonic characteristics of the onset component and the energy increases of the signal. Both properties are combined by utilizing FIR comb filters on a frame by frame basis in order to obtain an onset detection function, which is suitable for detecting slow onsets. The proposed approach improves upon existing methods in terms of the percentage of correct detections in signals containing slow onsets.

Using Virtual Reality to Enhance Electrical Safety and Design in the Built Environment

Electricity and the inherent risks associated with its use in the built environment have long since been a priority for the electrical services industry and also the general public who must live and work in this environment. By its nature virtual reality has the advantage of being safe for both the user and equipment. In addition, it offers the user an opportunity to be exposed to a range of scenarios and conditions that either occur infrequently or are hazardous to replicate. This paper presents a prototype desktop virtual reality model, to enhance electrical safety and design in the built environment. The model presented has the potential to be used as an educational tool for third level students, a design tool for industry, or as a virtual electrical safety manual for the general public. A description of the development of the virtual reality model is presented along with the applications that were developed within the model. The potential for virtual reality is highlighted with areas identified for future development. Based on the development of this prototype model, it appears that there is sufficient evidence to suggest that virtual reality could enhance electrical safety and design in the built environment and also advance training methods used to educate electrical services engineers and electricians.

Automating Ornamentation Transcription

A novel technique for detecting single and multi-note ornaments is presented. The system detects audio segments by utilising an onset detector based on comb filters (ODCF), which is capable of detecting very close events. In addition, a novel method to remove spurious onsets due to offset events is introduced. The system utilises musical ornamentation theory to decide whether a sequence of audio segments correspond to an ornamentation musical structure. In order to evaluate the results, a database of signals produced by different players using the three different instruments has been utilised. The results represent a step forward towards fully automating ornamentation transcription.

A Fiber Bragg Grating-Based All-Fiber Sensing System for Telerobotic Cutting Applications

A fiber Bragg grating (FBG)-based strain sensing system for minimally invasive telerobotic cutting applications is presented in this paper. Investigations assume that a scissor blade can be approximated as a uniformly tapered cantilever beam. A replica of the scissor blade is produced and strain characterization has been carried out using an FBG sensor system. Results are validated against measurements obtained using conventional electrical resistance strain gauges. The scissor blade experiences both direct and lateral forces during cutting, hence the system is characterized for a direct load range of 0-30 N and a lateral load range of 0-10 N. The results show a very good linear response for direct loading and some sensitivity to lateral loading. An actual sensorized scissor blade prototype is also characterized and results compared with that of the replica blade. The FBG interrogation system used was a macro-bend fiber filter-based ratiometric system. The use of FBGs together with macro-bend fiber-based interrogation system eliminates the influence of temperature on the sensing system and hence temperature independent strain information from the blade is obtained. The results obtained using the macro-bend fiber filter are compared with that of a commercial interrogation system and found to be in agreement. By implementing an all fiber sensing system based on fiber Bragg gratings and macrobend fiber filter interrogation system, remote operation of telerobotic cutting applications can be made more cost effective while providing a competitive accuracy and resolution solution.