Suggested reading: getting going with deep learning

Based on a conversation we had in the Machine Listening Lab last week, here are some blogs and other things you can read when you’re – say – a new PhD student who wants to get started with applying/understanding deep learning. We can recommend plenty of textbooks too, but here it’s mainly blogs and other informal introductions. Our recommended reading:

Andrew Ng’s coursera course on “Deep Learning”
– it’s not free to be a student on the course, BUT it is free to “audit” the course, either by signing in, or simply by watching the videos on Youtube

A brief overview of Deep Learning – a v good intro. (Also: DO read the comments. Some big names give their thoughts.)

This overview Nature paper is good too:

This blog post series covers the LINEAR ALGEBRA underlying deep learning and numerical optimisation

Introductions that show all the different types of NN architectures:

Deep learning book (free online)  by Ian Goodfellow and Yoshua Bengio and Aaron Courville

Deep learning book (free online) by Michael Nielsen


My Neural Network isn’t working! What should I do?
…you’ll need this!


a very readable tutorial on image generation using deep learning (specifically, GANs)

For Multi Task Learning:

For LSTMs (a popular type of recurrent neural network):

Posted in Uncategorized

Machine Listening Lab 2018: The year in review

2018 has been a fascinating year for the Machine Listening Lab. Here are the headlines!

Grant success and projects


  • MLLab Symposium 2018: featuring an invited industrial speaker (Katerina Kosta, Jukedeck), 3 internal speakers, and 11 lightning talks from students and postdocs. There were 42 attendees including MLLab members, interested people from around QMUL, and a collaborator visiting from Brazil (Rodrigo Schramm).
  • SoundCamp 2018: Animal Diplomacy Bureau, an art/design/public engagement commission by Kaylene Kau for Dan Stowell’s research project. Lots of members of the public, adults and children, engaged with bird sounds and birds’ lives in a park in South London. 43-ba8a4006-kaylene_explaining_kidsMLLab students helped make this a success: Will Wilkinson, Veronica Morfi, and Sophie McDonald.
  • QMUL Festival of Ideas, a group of MLLab academics and postdocs got together to play FolkRNN tunes from Bob Sturm’s algorithm.
  • The second Bird Audio Detection challenge took place in Summer 2018: a data challenge lead-organised by Dan Stowell. Thirteen teams from around the world took part, and the highest-scoring systems represented a dramatic improvement on the state of the art.
  • Emmanouil Benetos was Programme Co-Chair for the 19th International Society for Music Information Retrieval Conference (ISMIR 2018).
  • Machine Listening Lab researchers took part in the 2018 Workshop and Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE 2018) – presenting papers and also leading a task on Bird Audio Detection. (More info)
  • Dan Stowell announced a Special Session on “Wildlife Bioacoustics and Adaptive Signal Processing”, to happen at the ICASSP 2019 conference. He will chair it together with Naomi Harte and Theo Damoulas.


  • Dan Stowell and Emmanouil Benetos were awarded Turing Fellowships from The Alan Turing Institute (2018-2020).
  • Emmanouil Benetos received the “Outstanding Contribution to ISMIR Award” from the International Society for Music Information Retrieval (Oct 2018).
  • Emmanouil Benetos was promoted to Senior Lecturer (Oct 2018).
  • Emmanouil Benetos received a “Research Performance Award” from the Faculty of Science & Engineering of Queen Mary University of London (Jan 2018).

New people, visitors, and farewells


  • We said farewell to Bob Sturm – a founding member and co-leader of the MLLab, he has now moved to Stockholm to take up a position as Associate Professor at KTH.
  • Congratulations to MLLab PhD student Maria Panteli, who successfully defended her PhD on Computational Analysis of World Music Corpora (April 2018).

New people:

Visiting researchers:

  • Pavel Linhart and Tereza Petruskova (Czech Republic)
  • Vincent Lostanlen (Cornell University & New York University, USA)
  • Mathieu Lagrange and Felix Gontier (Ecole Centrale de Nantes, France)

Editorial activities

Invited talks

  • Dan Stowell gave the invited opening lecture at the 2018 Intelligent Sensing Summer School (video online here), and an invited talk at Silwood Park (Imperial College London).
  • Emmanouil Benetos gave a keynote on “Automatic transcription of world music collections” at the 8th International Workshop on Folk Music Analysis, Thessaloniki, Greece, June 2018.
  • Emmanouil Benetos gave an invited talk on “Automatic Music Transcription: Representations and Categorical (mis)Conceptions” at the 5th International Conference on Analytical Approaches to World Music, Thessaloniki, Greece, June 2018.

Selected Publications

Book chapters:

Approaches to complex sound scene analysis
E Benetos, D Stowell, and M D Plumbley
Computational Analysis of Sound Scenes and Events, Springer, 2018

Computational bioacoustic scene analysis
D Stowell
Computational Analysis of Sound Scenes and Events, Springer, 2018

Journal articles:

Deep Learning for Audio Event Detection and Tagging on Low-Resource Datasets
V Morfi and D Stowell
Applied Sciences 8 (8), 1397

Automatic acoustic detection of birds through deep learning: the first Bird Audio Detection challenge
D Stowell, MD Wood, H Pamuła, Y Stylianou, and H Glotin
Methods in Ecology and Evolution

A supervised classification approach for note tracking in polyphonic piano transcription
JJ Valero-Mas, E Benetos, and JM Iñesta
Journal of New Music Research, 1-15

A review of manual and computational approaches for the study of world music corpora
M Panteli, E Benetos, and S Dixon
Journal of New Music Research 47 (2), 176-189

Speaker recognition with hybrid features from a deep belief network
H Ali, SN Tran, E Benetos, and ASA Garcez
Neural Computing and Applications 29 (6), 13-19

Detection and classification of acoustic scenes and events: Outcome of the DCASE 2016 challenge
A Mesaros, T Heittola, E Benetos, P Foster, M Lagrange, T Virtanen, and MD Plumbley
IEEE/ACM Transactions on Audio, Speech and Language Processing (TASLP) 26 (2), 379-393

Posted in Events, Lab Updates, Publications

MLLab participates in DCASE 2018 Workshop and Challenge

Machine Listening Lab researchers will be participating at the 2018 Workshop and Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE). The workshop, which is at its third iteration, is taking place on 19-20 November 2018 in Surrey, UK, and aims to provide a venue for researchers working on computational sound scene analysis to present and discuss their work. The challenge, which is at its fourth edition, continues to support the development of computational sound scene analysis methods by comparing different approaches using common publicly available datasets.

The following papers by MLLab researchers will be presented at the DCASE 2018 Workshop:

In addition, Dan Stowell is the lead organiser of DCASE Challenge Task 3 which is all about bird audio detection. The challenge made use of data from collaborators all around the world, and 13 teams participated. Dan will be presenting the overall results of the bird audio detection task.

See you all in Surrey!

Posted in dcase, Events, Publications

PhD Studentships – Alan Turing Institute & QMUL

The Machine Listening Lab at Queen Mary University of London (QMUL) would like to invite applicants for PhD positions to be held jointly between the Alan Turing Institute and QMUL. Three PhD students across QMUL will be sponsored by this scheme, which provides 3.5-year studentships that come with an increased annual stipend of £20,500 plus a travel allowance and tuition fees. Doctoral students will split their time between QMUL and the Institute headquarters at the British Library in London. Students will be supervised by Turing fellows from QMUL and will work in collaboration with research teams at the Turing Institute.

The Machine Listening Lab is inviting applications for Turing-QMUL PhD projects in the following research topics:

  • Privacy-preserving urban/domestic sound monitoring
  • Automatic recognition of acoustic environments in sound archives
  • High-resolution machine listening for wildlife sounds

Please note that the Turing-QMUL PhD funding is across many subject areas and your application will be considered alongside applications from other disciplines. More information for the Turing-QMUL PhD positions is available at:

Candidates who are interested in carrying out research hosted by the Machine Listening Lab should apply directly to the School of Electronic Engineering and Computer Science of Queen Mary University of London at:

The deadline for submission is midday Monday 14 January 2019. In your application, and when choosing a supervisor, applicants should make it clear that they would like to be considered for the Turing doctoral studentship. Queen Mary will complete an initial assessment of your application and will refer selected candidates to the Institute in early February 2019. If successful, you will then be invited to attend a further interview at The Alan Turing Institute.

For informal queries before application, or to discuss topics, please email Dan Stowell and Emmanouil Benetos.

Posted in Opportunities

Special Session: “Wildlife Bioacoustics and Adaptive Signal Processing”, IEEE ICASSP 2019 (Brighton, UK)

– Special session at IEEE ICASSP 2019

Organisers: Dan Stowell, Naomi Harte, Theo Damoulas



Wildlife bioacoustics is witnessing a surge in both data volumes and computational methods. Monitoring projects worldwide collect many thousands of hours of audio each [1,2,3,4], and computational methods are now able to mine these datasets to detect, isolate and characterise recorded wildlife sounds at scale [2,5]. Such bioacoustic data is crucial for monitoring the rapid declines in many wildlife populations [6], as well as advancing the science of animal behaviour.

However, many open problems remain, including:

  • detecting/discriminating very brief low-SNR sounds;
  • estimating distance to animals from single-microphone recordings;
  • integrating evidence from multiple sensors, and from differing sensing sources (e.g. radar plus acoustics);
  • sampling bias, imbalance and concept drift in data sources;
  • high-resolution and perceptual measures of animal sound similarity;
  • sound source separation and localisation in complex wildlife sound scenes.

These problems must be solved in order to bring the data fully to bear on urgent global issues such as the loss of animal habitats.

New methods come from machine learning and advanced signal processing: matrix factorisations, deep learning, Gaussian processes, novel time-frequency transforms, and more. These new methods have been shown to be useful in various projects, across many species – bird, bat, cetacean and terrestrial mammals. However, much of this happens in isolated projects and there is a need to bring together lessons learned and establish state-of-the-art methods and directions for the future of this field.

The topic of computational wildlife bioacoustics is growing but lacks any dedicated conference or workshop. ICASSP is an ideal event through which to strengthen the application and the development of signal processing work in this increasingly important domain.



** Please note: submissions are not through the main ICASSP submission website! Use the link below. **

Full papers (4 pages, with an optional 5th page of references) should be submitted via the ICASSP special session submission system. The deadline is the same as for other ICASSP papers (Oct 29th 2018). The review process and the quality threshold for acceptance will also be the same.

For template and formatting information, please see the paper kit.



Organiser biographies:

Dan Stowell

Dan Stowell is Senior Researcher at Queen Mary University of London. He co-leads the Machine Listening Lab at Queen Mary University of London, based in the Centre for Digital Music, and is also a Turing Fellow at the Alan Turing Institute  Dan has worked on voice, music and environmental soundscapes, and is currently leading a five-year EPSRC fellowship project researching the automatic analysis of bird sounds.

Naomi Harte

Naomi Harte is Associate Professor of Digital Media Systems and a Fellow of Trinity College Dublin, Ireland. She is Co-PI of the ADAPT Research Centre, and a PI in the Sigmedia Research Group. Naomi’s primary focus is human speech communication, including speech quality, audio visual speech processing, speaker verification for biometrics and emotion in speech. Since 2012, she has collaborated on bird song analysis with the Dept. of Zoology in TCD.

Theo Damoulas

Theo Damoulas is an Associate Professor in Data Science with a joint appointment in Computer Science and Statistics at the University of Warwick. He is a Turing Fellow at the Alan Turing Institute and a visiting academic at NYU. His research interests are in machine learning and Bayesian statistics with a focus on spatio-temporal problems in urban science and computational sustainability.



[1] Lostanlen et al (2018) Birdvox-full-night: A Dataset and Benchmark for Avian Flight Call Detection In Proceedings of the IEEE International Conference on Acoustics , Speech, and Signal Processing (ICASSP), Calgary, Canada, Apr. 2018.

[2] Stowell et al (2018) “Automatic acoustic detection of birds through deep learning: the first Bird Audio Detection challenge.” Methods in Ecology and Evolution. In press.

[3] AmiBio project,

[4] Sullivan et al (2014), “The eBird enterprise: an integrated approach to development and application of citizen science.” Biological Conservation, vol. 169, pp. 31–40, 2014.

[5] Goeau et al (2016), “LifeCLEF Bird Identification Task 2016: The arrival of deep learning”, Working Notes of CLEF 2016, 440-449, 2016.

[6] Joppa (2017), “The case for technology investments in the environment.” Nature 552, 325-328, 2017.

Posted in Events

Turing Fellowship success for the Machine Listening Lab

Two of the Machine Listening Lab’s lead academics, Dan Stowell and Emmanouil Benetos, have been awarded Turing Fellowships.

The Alan Turing Institute (ATI) is a new UK national institute for data science and artificial intelligence (founded in 2015). Our university QMUL recently joined the ATI as a partner. The new QMUL Institute of Applied Data Sciences connects together QMUL’s researchers working on data science and artificial intelligence, and also acts as a conduit to the ATI.

Through the Turing Fellowships, effective 1st October 2018, Stowell and Benetos plan to work with ATI partners to build on their research themes in advanced audio analysis – to analyse urban, wildlife, domestic and musical sound recordings. They will work with academic, industry and government on the effective and ethical development of technology, including the development of privacy-preserving analysis methods.

Posted in Uncategorized

Bird audio detection DCASE challenge – less than a week to go

The second Bird Audio Detection challenge, now running as Task 3 in the 2018 DCASE Challenge, has been running all month, and the leaderboard is hotting up!

screenshot: leaderboard scores

There’s less than a week to go.

Who will get the strongest results? How well will these leaderboard preview scores predict the final outcome? (Note: the “preview” scores come from only 1000 audio files, from only two of our three evaluation sets. Whose algorithm generalises well?)

Who is this “ML” consistently getting preview scores above 90%? Will this beat the system from “Arjun” which was the first to get past the 90% mark? Will someone get a final score above 90%? Above 95%?

These scores aren’t just for fun, they represent (indirectly) the amount of manual labour saved by an automatic detector. If the AUC score gets twice as close to the 100% mark, this indicates approximately halving the number of false-positive detections that you have to tolerate from your system, which can thus save hundreds of person-hours in data mining, or enable some automation that wasn’t possible before.

Also importantly, what are the advances in the state of the art that are bringing these high-quality results? In the first Bird Audio Detection challenge, which was only last year, the strongest system attained 88.7% AUC – and this time we made the task harder by expanding to more different datasets, and by using a slightly modified scoring measure. (Instead of the overall AUC, we use the harmonic mean of the AUCs obtained with each of the three evaluation datasets. This tends to yield lower scores, especially for systems which do well at some datasets and poorly at others.) Machine learning is moving fast, and it’s not always clear which new developments provide real benefits in practical applications. It’s clear that some innovations have come along since last year, for bird audio detection.

So, we wait with interest to see the technical reports that will be submitted to the DCASE workshop (deadline next week, 31st July!)

Posted in Bird, dcase

SoundCamp 2018 – Animal Diplomacy Bureau

Over the recent holiday weekend, in order to spread ideas about bird sounds and ecology, we made groups of kids and grown-ups run around in a field wearing bird heads and searching for food.


This was Animal Diplomacy Bureau, a game created by Kaylene Kau. Since my research fellowship is all about birds and sounds and how they interact, and the game is all about exploring those topics, I was really pleased to commission Kaylene to develop an expanded version of the game and show it as part of SoundCamp 2018.

It was a glorious sunny bank holiday weekend, and the game was really popular, kids and grown-ups queueing up to play. They had to find food tokens in the park while either hunting or being hunted by other species, and using the sound of bird calls as clues.


The game consisted of participants taking on the role of either a parakeet, a goldfinch or a sparrowhawk while searching for red discs representing berries. The food location was indicated by recorded goldfinch calls heard nearby. The sparrowhawks had to wear small loudspeakers, which would play back sparrowhawk alarm calls giving the prey a chance to react. The prey players could resist the sparrowhawks by getting together and “mobbing” the bird, which is what many birds do in real life.

After the game, Kaylene hosted discussions where she talked about how the bird species they’d been playing interact in real life, and how living in the city affects them. She invited participants to discuss what cities would be like if they were designed for animals as well as for humans.


We had dozens of players, and also dozens of queries from people in the park curious to know what was going on with these bird people foraging around the place. Participants went away with an inside perspective on how it is to be a goldfinch, a sparrowhawk or a parakeet in London!

Thanks to my PhD students for helping Kaylene to run the event: Will Wilkinson, Veronica Morfi and Sophie McDonald.

(Financial support: EPSRC Fellowship EP/L020505/1 (Dan Stowell))


Posted in Bird, Events

Machine Listening Lab symposium 2018

On 29th March 2018 we held a Machine Listening Lab symposium to gather together people across the university who are doing work in machine listening.



Katerina Kosta (Jukedeck)

The programme of talks included:

  • Emmanouil Benetos (QMUL):
    Machine listening for music and everyday sounds: the year ahead
  • Rob Lachlan (QMUL):
    Learning about bird song learning via inference from population-level variation
  • Katerina Kosta (invited guest speaker):
    Creating music at Jukedeck
  • Michael McLoughlin (QMUL):
    Sea to Farm: Bioacoustics in Animal Behaviour and Welfare

Plus 12 lightning talks from students and postdocs, and from Rodrigo Schramm visiting from Brazil to update us on his work since spending time with us last year.


Rodrigo Schramm (UFRGS)

Thank you to everyone who took part!



(Funded by Dan Stowell EPSRC Research Fellowship EP/L020505/1)

Posted in Uncategorized

Many C4DM papers accepted for ICASSP 2018

C4DM researchers have had a lot of success this year in being accepted for ICASSP 2018, the IEEE’s International Conference on Acoustics, Speech and Signal Processing. Most of these papers are led by C4DM PhD students, on MIR and Machine Listening topics:

  • “A Deeper Look At Gaussian Mixture Model Based Anti-Spoofing Systems” by Bhusan Chettri and Bob L. Sturm
  • “Towards Complete Polyphonic Music Transcription: Integrating Multi-Pitch Detection and Rhythm Quantization” by Eita Nakamura, Emmanouil Benetos, Kazuyoshi Yoshii, and Simon Dixon
  • “Polyphonic Music Sequence Transduction With Meter-Constrained LSTM Networks” by Adrien Ycart and Emmanouil Benetos
  • “Feature Design Using Audio Decomposition for Intelligent Control of the Dynamic Range Compressor” by Di Sheng and György Fazekas
  • Adversarial Semi-Supervised Audio Source Separation applied to Singing Voice Extraction” by Daniel Stoller, Sebastian Ewert, Simon Dixon
  • “Similarity Measures For Vocal-Based Drum Sample Retrieval Using Deep Convolutional Auto-Encoders” by Adib Mehrabi, Kuenwoo Choi, Simon Dixon, Mark Sandler
  • Shift-Invariant Kernel Additive Modelling for Audio Source Separation” by Delia Fano Yela, Sebastian Ewert, Ken O’Hanlon, Mark B. Sandler
  • “Improved detection of semi-percussive onsets in audio using temporal reassignment” by K. O’Hanlon and M.B. Sandler

See you all in Calgary!

Posted in Publications