Music and Speech Modeling 2018: Final Projects

Really proud of my students in the postgraduate class on Music and Speech Modeling (ECS792) who again rose to the occasion to conceive, design, and develop creative final projects on topics related to music performance and speech prosody — see the list of Music and Speech Modeling Final Projects. The students were PhD students from the Media and Arts Technology Doctoral Training Centre and MSc students from the Sound and Music Computing programme at QMUL.

We had two speech-related projects. Louise Bryce compared famous lines by their original speakers and by impersonators, while Guillermo Peeters designed an experiment to evaluate whether familiarity with a foreign language speaker makes the foreign speech of that speaker more comprehensible.

The Impersonation Game Louise Bryce Many actors have won awards for their impersonations of real people in film and television. This project investigates the speech qualities in four of these impersonations—Meryl Streep playing Margaret Thatcher, Natalie Portman as Jackie Kennedy, James Franco as Tommy Wiseau, Colin Firth as King George VI—to discuss the differences and similarities in each performance when compared to the real person saying the same phrase.

Effects of prosody and accommodation on the comprehension of an unfamiliar language Guillermo Peeters We tend to adapt to the patterns of stress, rhythm and intonation of our close relatives and our significant others. This project investigates to which extent prosody and vocal accommodation facilitate the inference of meaning cross-linguistically on dyadic relationships.

In the realm of music performance, David Foster mapped the amplitude and pitch variations of every bass note sample to create a dictionary to approximate natural performances, Amogh Matt measured changes to the heavy metal singer James Hetfield's voice over the years, and James Weaver quantified the effect of spatial acoustics on performed tempo.

	88x127: Deconstructing a sample library David Foster It is difficult to get a life-like performance from sample based virtual instruments without expert tweaking of parameters. This project produced a tool which shows the user the amplitude and pitch variations of every sample, so that they can more accurately reproduce or create a musical performance.
	Analysis of the evolution of the singing voice over the years. Amogh Kabbinakanthi Matt Amongst the fans of the heavy metal band Metallica, James Hetfield's vocal transition after the ‘Justice’ album is often a topic of discussion. This project presents the evolution of his vocal characteristics throughout his career by analysing the track ‘Nothing Else Matters’ as recorded yearly from 1993 through 2018.
	Analysing the impact of reverberation on expressivity in ensemble performance James Weaver This study looks at how varying reverberation conditions on two performers playing a duet changes their expressivity. This project looks specifically at tempo, and loudness change from performance to performance, and builds a replicable methodology for future studies and analysis of ensemble performance.

Related to music perception, Edward Hall computed the entropy of different surface features to predict the saliency of different music features, while Georgios Kefalas assessed the impact of ambient sound on time perception.

	Modelling the Musical Surface Edward Hall This project attempts to build a computational model of the musical surface – the most salient features within the texture that we actually pay attention to when listening to music. With this model, it is hoped that we can understand better the ways in which we perceive music.
	Time estimation through ambient soundscape Georgios Kefalas For listeners, temporal awareness seems to be lessened while listening to ambient music. This study tries to examine how accurately people can estimate time through an ambient soundscape.