This Message Is From an External Sender
This message came from outside your organization.
Hello!
Singing synthesis is not my field but I did work at MTG for a little while during this period. Xavier was my boss and Jordi and Alex ran the group up the hall. My recollection was that they argued that in a number of impulse-train-generated musical timbres, it was important not only to reproduce the harmonic content but also to reproduce an impulse preserving phase aligment (the example I recall was violin, where the stick-slip transitions formed the impulse instants). Pretty sure there are impulse-aligned time domain time stretching methods known to work well on voice too. I have not read [2]. Do they give practical methods for recovering glottal closure instants from time domain (or frequency domain) audio signals? What parameters are you optimising? What is the cost function? Best, Ross. On 26/07/2026 7:42 am, QueueSevenM wrote: > Hello, I am posting my introduction here as recommended on the joining > instructions page. Unlike presumably most of the people here, I am not > an academic or professional researcher. I have been learning DSP in my > spare time over the course of > ZjQcmQRYFpfptBannerStart > This Message Is From an External Sender > This message came from outside your organization. > ZjQcmQRYFpfptBannerEnd > > Hello, > > I am posting my introduction here as recommended on the joining > instructions page. Unlike presumably most of the people here, I am not > an academic or professional researcher. I have been learning DSP in my > spare time over the course of the last several months. What I am mainly > joining to post about is research I have been conducting on improvements > to various audio processing techniques and models, specifically with > respect to voice processing and even more specifically, singing voice > synthesis. > > I am currently undertaking the following: > - A system which is able to automatically find optimal values to the > various parameters and constants within configurations that are used by > various audio processing algorithms; avoiding the tedium of manually > determining them and hopefully getting better results too. > - A method for modeling the correlation between the phases in the > Spectral Modeling Synthesis [1] residual and glottal closure instants. > [2] (which in turn cites [3]) notes the importance of this. [4] notes > that SMS performs particularly poorly in voiced fricatives because of > the loss of this correlation. > - Improvements to the Excitation plus Resonance [5] model. > > In the future, I will also work on improvements to Wide-Band Harmonic > Sinusoidal Modeling [6]. > > References: > [1] Serra, Xavier. "A System for Sound > Analysis/Transformation/Synthesis based on a Deterministic plus > Stochastic Decomposition". Stanford University, PhD Dissertation, 1989. > [2] Bonada Sanjaume, Jordi; Serra, Xavier. "Synthesis of the Singing > Voice by Performance Sampling and Spectral Models," in IEEE Signal > Processing Magazine, vol. 24, no. 2, pp. 67-79, March 2007. > [3] Kob, Malte. "Physical Modeling of the Singing Voice". Institute of > Technical Acoustics, > Aachen University, PhD dissertation, 2002. > [4] Bonada Sanjaume, Jordi; Loscos, Alex; Mayor, Oscar; Kenmochi, > Hideki. "Sample-Based Singing Voice Synthesizer using Spectral Models > and Source-Filter Decomposition". Proceedings of Interspeech 2003. > [5] Bonada Sanjaume, Jordi; Celma, Oscar; Loscos, Alex; Ortola, Jaume; > Serra, Xavier. "Singing Voice Synthesis Combining Excitation plus > Resonance and Sinusoidal plus Residual Models". Pompeu Fabra University, > 2001. >
