All posts by Christos Plachouras

On 31 August to 5 September, AIM PhD students will participate in Interspeech 2024 and its satellite events. Interspeech is the premier international conference for research on the science and technology of spoken language processing.

Chin-Yun Yu will present his paper “Differentiable Time-Varying Linear Prediction in the Context of End-to-End Analysis-by-Synthesis“. The paper introduces improvements to the GOLF voice synthesizer, by implementing a joint filtering approach for noise and harmonics using a single LP (Linear Prediction) filter, resembling a classic source-filter model, and replacing frame-wise approximation with sample-by-sample LP processing, implemented efficiently in C++ and CUDA. These modifications result in smoother spectral envelopes, reduced artefacts, and improved performance in listening tests compared to other baselines. More information can be found here.

Farida Yusuf is part of the programme committee for the Young Female Researchers in Speech Workshop (YFRSW). The workshop is designed for Bachelor’s and Master’s students currently engaged in speech science and technology research, aiming to promote interest in the field among those who haven’t yet committed to pursuing a PhD. It features panel discussions, student poster presentations, and mentoring sessions, providing participants with opportunities to showcase their research and engage with PhD students and senior researchers in the field.

See you at Interspeech!

On 2-6 September, AIM PhD students Jordie Shier, Shuoyang Zeng, and Teresa Pelinski will be at the International Conference on New Interfaces for Musical Expression (NIME), which will take place in Utrecht, the Netherlands.

Jordie will present his paper Real-time Timbre Remapping with Differentiable DSP, written in collaboration with Charalampos Saitis (C4DM, QMUL), Andrew Robertson (Ableton) and Andrew McPherson (Imperial College London). The paper discusses a method for mapping audio from percussion instruments onto synthesiser controls in real-time using neural networks, enabling nuanced and audio-driven timbral control of a musical synthesiser. You can read the paper here and check the project website and presentation here.

Shuoyang will also present a paper, Building sketch-to-sound mapping with unsupervised feature extraction and interactive machine learning, written in collaboration with AIM PhD student Bleiz Del Sette and Charalampos Saitis (C4DM, QMUL), Anna Xambó (C4DM, QMUL) and Nick Bryan-Kinns (CCI, University of the Arts London). The paper explores interactive (personalised) constructions of mappings between visual sketches and sound controls as an expressive way for musical composition and performance.

Teresa will co-lead two workshops. The first workshop, First- and second-person perspectives for ML in NIME has been organised in collaboration with Théo Jourdan (Sorbonne Université) and Hugo Scurto (independent artist and researcher). It focuses on autoethnographic methods to articulate insights and experiences surrounding new instrument building with AI – you can read more here. The second workshop, Building NIMEs with Embedded AI, has been organised in collaboration with Charles Patrick Martin (Australian National University). It is a hands-on tutorial for embedding light deep learning models on Raspberry Pi and Bela – you can read more here.

See you at NIME!

On 26-30 August, AIM researchers will participate in the 32nd European Signal Processing Conference (EUSIPCO 2024) in Lyon, France. EUSIPCO is the flagship conference of the European Signal Processing Society (EURASIP) and offers a comprehensive technical program addressing all the latest developments in research and technology for signal processing.

AIM members will be presenting the following work:

• Towards building an end-to-end multilingual automatic lyrics transcription model, by Jiawen Huang and Emmanouil Benetos

See you at EUSIPCO!