Search

Previous research has shown that established techniques for spoken voice conversion (VC) do not perform as well when applied to singing voice conversion (SVC). We propose an alternative loss component in a loss function that is otherwise well-established among VC tasks, which has been shown to improve our model's SVC performance. We first trained a singer identity embedding (SIE) network on mel-spectrograms of singer recordings to produce singer-specific variance encodings using contrastive learning. We subsequently trained a well-known autoencoder framework (AutoVC) conditioned on these SIEs, and measured differences in SVC performance when using different latent regressor loss components. We found that using this loss w.r.t. SIEs leads to better performance than w.r.t. bottleneck embeddings, where converted audio is more natural and specific towards target singers. The inclusion of this loss component has the advantage of explicitly forcing the network to reconstruct with timbral similarity, and also negates the effect of poor disentanglement in AutoVC's bottleneck embeddings. We demonstrate peculiar diversity between computational and human evaluations on singer-converted audio clips, which highlights the necessity of both. We also propose a pitch-matching mechanism between source and target singers to ensure these evaluations are not influenced by differences in pitch register., Comment: Submitted to the Sound and Music Computing Conference 2023

This paper makes several contributions to automatic lyrics transcription (ALT) research. Our main contribution is a novel variant of the Multistreaming Time-Delay Neural Network (MTDNN) architecture, called MSTRE-Net, which processes the temporal information using multiple streams in parallel with varying resolutions keeping the network more compact, and thus with a faster inference and an improved recognition rate than having identical TDNN streams. In addition, two novel preprocessing steps prior to training the acoustic model are proposed. First, we suggest using recordings from both monophonic and polyphonic domains during training the acoustic model. Second, we tag monophonic and polyphonic recordings with distinct labels for discriminating non-vocal silence and music instances during alignment. Moreover, we present a new test set with a considerably larger size and a higher musical variability compared to the existing datasets used in ALT literature, while maintaining the gender balance of the singers. Our best performing model sets the state-of-the-art in lyrics transcription by a large margin. For reproducibility, we publicly share the identifiers to retrieve the data used in this paper.

The Filosax dataset is a large collection of specially commissioned recordings of jazz saxophonists playing with commercially available backing tracks. Five participants each recorded themselves playing the melody, interpreting a transcribed solo and improvising on 48 tracks, giving a total of around 24 hours of audio data. The solos are annotated both as individual note events with physical timing, and as sheet music with a metrical interpretation of the timing. In this paper, we outline the criteria used for choosing and sourcing the repertoire, the recording process and the semi-automatic transcription pipeline. We demonstrate the use of the dataset to analyse musical phenomena such as swing timing and dynamics of typical musical figures, as well as for training a source activity detection system and predicting expressive characteristics. Other potential applications include the modelling of jazz improvisation, performer identification, automatic music transcription, source separation and music generation.

IMPORTANCE: Robot-assisted radical cystectomy is being performed with increasing frequency, but it is unclear whether total intracorporeal surgery improves recovery compared with open radical cystectomy for bladder cancer. OBJECTIVES: To compare recovery and morbidity after robot-assisted radical cystectomy with intracorporeal reconstruction vs open radical cystectomy. DESIGN, SETTING, AND PARTICIPANTS: Randomized clinical trial of patients with nonmetastatic bladder cancer recruited at 9 sites in the UK, from March 2017-March 2020. Follow-up was conducted at 90 days, 6 months, and 12 months, with final follow-up on September 23, 2021. INTERVENTIONS: Participants were randomized to receive robot-assisted radical cystectomy with intracorporeal reconstruction (n = 169) or open radical cystectomy (n = 169). MAIN OUTCOMES AND MEASURES: The primary outcome was the number of days alive and out of the hospital within 90 days of surgery. There were 20 secondary outcomes, including complications, quality of life, disability, stamina, activity levels, and survival. Analyses were adjusted for the type of diversion and center. RESULTS: Among 338 randomized participants, 317 underwent radical cystectomy (mean age, 69 years; 67 women [21%]; 107 [34%] received neoadjuvant chemotherapy; 282 [89%] underwent ileal conduit reconstruction); the primary outcome was analyzed in 305 (96%). The median number of days alive and out of the hospital within 90 days of surgery was 82 (IQR, 76-84) for patients undergoing robotic surgery vs 80 (IQR, 72-83) for open surgery (adjusted difference, 2.2 days [95% CI, 0.50-3.85]; P = .01). Thromboembolic complications (1.9% vs 8.3%; difference, –6.5% [95% CI, –11.4% to –1.4%]) and wound complications (5.6% vs 16.0%; difference, –11.7% [95% CI, –18.6% to –4.6%]) were less common with robotic surgery than open surgery. Participants undergoing open surgery reported worse quality of life vs robotic surgery at 5 weeks (difference in mean European Quality of Life 5-Dimension, 5-Level instrument scores, –0.07 [95% CI, –0.11 to –0.03]; P = .003) and greater disability at 5 weeks (difference in World Health Organization Disability Assessment Schedule 2.0 scores, 0.48 [95% CI, 0.15-0.73]; P = .003) and at 12 weeks (difference in WHODAS 2.0 scores, 0.38 [95% CI, 0.09-0.68]; P = .01); the differences were not significant after 12 weeks. There were no statistically significant differences in cancer recurrence (29/161 [18%] vs 25/156 [16%] after robotic and open surgery, respectively) and overall mortality (23/161 [14.3%] vs 23/156 [14.7%]), respectively) at median follow-up of 18.4 months (IQR, 12.8-21.1). CONCLUSIONS AND RELEVANCE: Among patients with nonmetastatic bladder cancer undergoing radical cystectomy, treatment with robot-assisted radical cystectomy with intracorporeal urinary diversion vs open radical cystectomy resulted in a statistically significant increase in days alive and out of the hospital over 90 days. However, the clinical importance of these findings remains uncertain. TRIAL REGISTRATION: ISRCTN Identifier: ISRCTN13680280; ClinicalTrials.gov Identifier: NCT03049410

This paper proposes a deep convolutional neural network for performing note-level instrument assignment. Given a polyphonic multi-instrumental music signal along with its ground truth or predicted notes, the objective is to assign an instrumental source for each note. This problem is addressed as a pitch-informed classification task where each note is analysed individually. We also propose to utilise several kernel shapes in the convolutional layers in order to facilitate learning of efficient timbre-discriminative feature maps. Experiments on the MusicNet dataset using 7 instrument classes show that our approach is able to achieve an average F-score of 0.904 when the original multi-pitch annotations are used as the pitch information for the system, and that it also excels if the note information is provided using third-party multi-pitch estimation algorithms. We also include ablation studies investigating the effects of the use of multiple kernel shapes and comparing different input representations for the audio and the note-related information., Comment: 4 figures, 4 tables and 7 pages. Accepted for publication at ISMIR Conference 2021

Sixty participants provided dissimilarity ratings between various singing techniques. Multidimensional scaling, class averaging and clustering techniques were used to analyse timbral spaces and how they change between different singers, genders and registers. Clustering analysis showed that ground-truth similarity and silhouette scores that were not significantly different between gender or register conditions, while similarity scores were positively correlated with participants' instrumental abilities and task comprehension. Participant feedback showed how a revised study design might mitigate noise in our data, leading to more detailed statistical results. Timbre maps and class distance analysis showed us which singing techniques remained similar to one another across gender and register conditions. This research provides insight into how the timbre space of singing changes under different conditions, highlights the subjectivity of perception between participants, and provides generalised timbre maps for regularisation in machine learning., In Proceedings of the 2020 Joint Conference on AI Music Creativity (CSMC-MuMe 2020), Stockholm, Sweden, October 15-19, 2020

In 2010 NICE released a clinical guideline recommending that natriuretic peptide (NP) testing in patients with suspected heart failure without previous myocardial infarction can accelerate diagnosis of heart failure and also avoid unnecessary echocardiography. A framework for the evaluation of the value of implementation activities is applied to this recommendation for NP testing.

Additional file 2. Unit costs applied for valuation of resource use.

Additional file 5. Other reasons given for not wanting to attend SSME if invited.

In 2010 NICE released a clinical guideline recommending that natriuretic peptide (NP) testing in patients withsuspected heart failure without previous myocardial infarction can accelerate diagnosis of heart failure and alsoavoid unnecessary echocardiography. A framework for the evaluation of the value of implementation activities isapplied to this recommendation for NP testing.

Additional file 7. Summary of the findings for each feasibility objective.

Additional file 1. Ethnographic interview topic guides.

Additional file 4. Other reasons given for not attending SSME.

Audio-to-score alignment aims at generating an accurate mapping between a performance audio and the score of a given piece. Standard alignment methods are based on Dynamic Time Warping (DTW) and employ handcrafted features. We explore the usage of neural networks as a preprocessing step for DTW-based automatic alignment methods. Experiments on music data from different acoustic conditions demonstrate that this method generates robust alignments whilst being adaptable at the same time., Comment: ML4MD at ICML 2019

Additional file 3. Summary of extracted continuous primary care data by practice.

Additional file 6. Case examples demonstrating impact.

Additional file 1.

Economic evaluation combines information on costs and benefits to inform priority setting in health care and to inform decisions on the reimbursement of health care interventions. Costs are typically the direct costs of providing health care but can also include indirect costs which occur as a result of having poor health. One such indirect cost is the cost associated with lost productivity in both paid and unpaid work. The inclusion of productivity costs in economic evaluation has the potential to have an impact on the decisions on whether or not to recommend or fund an intervention. Productivity loss has been measured by asking patients to report the time off work/normal activities due to poor health. However, this information is not collected in all trials or observational studies. A different approach would be to predict the number of days off work/normal activities associated with different levels of health related quality of life (HRQoL). This approach has been undertaken using Dutch respondents based on hypothetical health states (Krol et al, 2013) but not UK patient data.

The DAFNE structured education programme for adults with Type 1 diabetes is traditionally delivered over five consecutive days. However, it is not always convenient for individuals to find the time to attend an intensive five-day programme due to work and other obligations.This aim of this study was to provide a preliminary estimate of the cost-effectiveness of Dose Adjustment For Normal Eating (DAFNE) structured education for adults with Type 1 diabetes delivered one day per week over five weeks (‘5-week DAFNE’) compared with standard DAFNE (five consecutive days; ‘1-week DAFNE’).

Generative adversarial networks (GANs) have shown great success in applications such as image generation and inpainting. However, they typically require large datasets, which are often not available, especially in the context of prediction tasks such as image segmentation that require labels. Therefore, methods such as the CycleGAN use more easily available unlabelled data, but do not offer a way to leverage additional labelled data for improved performance. To address this shortcoming, we show how to factorise the joint data distribution into a set of lower-dimensional distributions along with their dependencies. This allows splitting the discriminator in a GAN into multiple "sub-discriminators" that can be independently trained from incomplete observations. Their outputs can be combined to estimate the density ratio between the joint real and the generator distribution, which enables training generators as in the original GAN framework. We apply our method to image generation, image segmentation and audio source separation, and obtain improved performance over a standard GAN when additional incomplete training examples are available. For the Cityscapes segmentation task in particular, our method also improves accuracy by an absolute 14.9% over CycleGAN while using only 25 additional paired examples., Comment: 10 pages plus 14 pages appendix. Accepted at the International Conference on Learning Representations (ICLR) 2020. Camera-ready submission. Implementation available at https://github.com/f90/FactorGAN

This paper presents two novel user interfaces for investigating the pattern content in monophonic jazz solos and exemplifies how these interfaces could be used for research on jazz improvisation. In jazz improvisation, patterns are of particular interest for the analysis of improvisation styles, the oral transmission of musical language, the practice of improvisation, and the psychology of creative processes. The ongoing project "Dig That Lick" is devoted to addressing these questions with the help of a large database of jazz solo transcriptions generated by automated melody extraction algorithms. To expose these transcriptions to jazz researchers, two prototypes of user interfaces were designed that work currently with the 456 manually transcribed jazz solos of the Weimar Jazz Database. The first one is a Shiny application that allows exploring a set of 653 of the most common patterns by eminent players. The second one is a web interface for a general two-staged pattern search in the Weimar Jazz Database featuring regular expressions. These applications aim on the one hand at an expert audience of jazz researchers to facilitate generating and testing hypotheses about patterns in jazz improvisation, and on the other hand at a wider audience of jazz teachers, students, and fans.

Methods for interpreting machine learning models can help one understand their global and/or local behaviours, and thereby improve them. In this work, we apply a global analysis method to a machine listening model, which essentially inverts the features generated in a model back into an interpretable form like a sonogram. We demonstrate this method for a state-of-the-art singing voice detection model. We train up-convolutional neural networks to invert the feature generated at each layer of the model. The results suggest that the deepest fully connected layer of the model does not preserve temporal and harmonic structures, but that the inverted features from the deepest convolutional layer do. Moreover, a qualitative analysis of a large number of inputs suggests that the deepest layer in the model learns a decision function as the information it preserves depends on the class label associated with an input.

Supplemental material, Supplementary_Material for Investigating style evolution of Western classical music: A computational approach by Christof Weiß, Matthias Mauch, Simon Dixon and Meinard Müller in Musicae Scientiae

Large wood in rivers and logjams are linked to the presence of varied riverine morphologies and increased abundance and diversity of aquatic biota. Current research into the ecohydrological, morphological and geochemical effects of logjams is restricted by difficulties in comparing findings between river systems. The problem is exacerbated by a lack of standardised metrics for recording and reporting logjams and a scale dependence of logjam effects with river size.In this paper a new method for analysing logjams is presented based on a set of seven dimensionless metrics of structure and morphological effects. These metrics are used along with a cluster analysis to identify key logjam types within a study river. The analysis framework is applied to data from a small forest river in the UK and identifies that 73% of logjams in this system can be grouped into eight distinct classes. Of these classes, two are logjams which only partly fill the channel and six are channel spanning. The individual classes are differentiated from each other principally by the degree of lateral and vertical erosion found in association with the logjams.The dimensionless metrics are also applied qualitatively to a range of logjams described in the literature and show the potential for the method to be applied as a standard survey and analysisframework for logjams across diverse river environments. The potential to link logjam form with function, and therefore provision of specific habitats, has wide applicability to the design and monitoring of river restoration schemes.

River channel confluences are widely acknowledged as important geomorphological nodes that control the downstream routing of water and sediment, and which are locations for the preservation of thick fluvial deposits overlying a basal scour. Despite their importance, there has been little study of the stratigraphic characteristics of river junctions, or the role of confluence morphodynamics in influencing stratigraphic character and preservation potential. As a result, although it is known that confluences can migrate through time, models of confluence geomorphology and sedimentology are usually presented from the perspective that the confluence remains at a fixed location. This is problematic for a number of reasons, not least of which is the continuing debate over whether it is possible to discriminate between scour that has been generated by autocyclic processes (such as confluence scour) and that driven by allocyclic controls (such as sea-level change). This paper investigates the spatial mobility of river confluences by using the 40-year record of Landsat Imagery to elucidate the styles, rates of change and areal extent over which large river confluence scours may migrate. On the basis of these observations, a new classification of the types of confluence scour is proposed and applied to the Amazon and Ganges-Brahmaputra-Meghna (GBM) basins. This analysis demonstrates that the2drivers of confluence mobility are broadly the same as those that drive channel change more generally. Thus in the GBM basin, a high sediment supply, large variability in monsoonal driven discharge and easily erodible bank materials result in a catchment where over 80 % of large confluences are mobile over this 40-year window; conversely this figure is less than 40 % for the Amazon basin. These results highlight that: i) the potential areal extent of confluence scours is much greater than previously assumed, with the location of some confluences on the Jamuna (Brahmaputra) River migrating over a distance of 20 times the tributary channel width; ii) extensive migration in the confluence location is more common than currently assumed, and iii) confluence mobility is often tied to the lithological and hydrological characteristics of the drainage basins that determine sediment yield.

[TODO] Add abstract here.

[TODO] Add abstract here.

Large music content libraries often comprise multiple versions of a piece of music. To establish a link between different versions, automatic music alignment methods map each position in one version to a corresponding position in another version. Due to the leeway in interpreting a piece, any two versions can differ significantly, for example, in terms of local tempo, articulation, or playing style. For a given pair of versions, these differences can be significant such that even state-of-the-art methods fail to identify a correct alignment. In this paper, we present a novel method that increases the robustness for difficult to align cases. Instead of aligning only pairs of versions as done in previous methods, our method aligns multiple versions in a joint manner. This way, the alignment can be computed by comparing each version not only with one but with several versions, which stabilizes the comparison and leads to an increase in alignment robustness. Using recordings from the Mazurka Project, the alignment error for our proposed method was 14% lower on average compared to a state-of-the-art method, with significantly less outliers (standard deviation 53% lower)., Comment: International Society for Music Information Retrieval Conference (ISMIR)

[TODO] Add abstract here.

This paper outlines a theoretical framework for creative technology based on two contrasting processes: divergent exploration and convergent optimisation. We claim that these two cases require different gesture-to-parameter mapping properties. Results are presented from a user experiment that motivates this theory. The experiment was conducted using a publicly available iPad app: ``Sonic Zoom''. Participants were encouraged to conduct an open ended exploration of synthesis timbre using a combination of two different interfaces. The first was a standard interface with ten sliders, hypothesised to be suited to the ``convergent'' stage of creation. The second was a mapping of the entire 10-D combinatorial space to a 2-D surface using a space filling curve. This novel interface was intended to support the ``divergent'' aspect of creativity. The paths of around 250 users through both 2-D and 10-D space were logged and analysed. Both the interaction data and questionnaire results show that the different interfaces tended to be used for different aspects of sound creation, and a combination of these two navigation styles was deemed to be more useful than either individually. The study indicates that the predictable, separate parameters found in most music technology are more appropriate for convergent tasks.

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

Publication in the conference proceedings of EUSIPCO, Bucharest, Romania, 2012

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

[TODO] Add abstract here.

&nbsp

[TODO] Add abstract here.