Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Singing Voice Synthesis Using Differentiable LPC and Glottal-Flow-Inspired Wavetables

Chin-Yun Yu, György Fazekas

TL;DR

It is shown that GOLF can model the phase components of the human voice, which has immense potential for rendering and analysing singing voices in a differentiable manner and underscore the advantages of signal-processing-based approaches, which offer greater interpretability and efficiency in synthesis.

Abstract

This paper introduces GlOttal-flow LPC Filter (GOLF), a novel method for singing voice synthesis (SVS) that exploits the physical characteristics of the human voice using differentiable digital signal processing. GOLF employs a glottal model as the harmonic source and IIR filters to simulate the vocal tract, resulting in an interpretable and efficient approach. We show it is competitive with state-of-the-art singing voice vocoders, requiring fewer synthesis parameters and less memory to train, and runs an order of magnitude faster for inference. Additionally, we demonstrate that GOLF can model the phase components of the human voice, which has immense potential for rendering and analysing singing voices in a differentiable manner. Our results highlight the effectiveness of incorporating the physical properties of the human voice mechanism into SVS and underscore the advantages of signal-processing-based approaches, which offer greater interpretability and efficiency in synthesis.

Singing Voice Synthesis Using Differentiable LPC and Glottal-Flow-Inspired Wavetables

TL;DR

It is shown that GOLF can model the phase components of the human voice, which has immense potential for rendering and analysing singing voices in a differentiable manner and underscore the advantages of signal-processing-based approaches, which offer greater interpretability and efficiency in synthesis.

Abstract

This paper introduces GlOttal-flow LPC Filter (GOLF), a novel method for singing voice synthesis (SVS) that exploits the physical characteristics of the human voice using differentiable digital signal processing. GOLF employs a glottal model as the harmonic source and IIR filters to simulate the vocal tract, resulting in an interpretable and efficient approach. We show it is competitive with state-of-the-art singing voice vocoders, requiring fewer synthesis parameters and less memory to train, and runs an order of magnitude faster for inference. Additionally, we demonstrate that GOLF can model the phase components of the human voice, which has immense potential for rendering and analysing singing voices in a differentiable manner. Our results highlight the effectiveness of incorporating the physical properties of the human voice mechanism into SVS and underscore the advantages of signal-processing-based approaches, which offer greater interpretability and efficiency in synthesis.
Paper Structure (22 sections, 10 equations, 5 figures, 2 tables)

This paper contains 22 sections, 10 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Glottal Source-Filter Model
  4. Linear Predictive Coding
  5. Proposed Model
  6. Glottal Flow Wavetables
  7. Frame-Wise LPC Synthesis
  8. LPC Coefficients Parameterisation
  9. Unvoiced Gating
  10. Optimisation
  11. Backpropagation Through the Coefficients
  12. Backpropagation Through the Input
  13. Experimental Setup
  14. Dataset
  15. Model Details
  16. ...and 7 more sections

Figures (5)

  • Figure 1: An example of the wavetables we used, corresponding to matrix $\mathbf{D}$ with $K = 31$.
  • Figure 2: Overview of the GOLF synthesis process. phase offset is only introduced at test time, where the details are given in Section \ref{['ssec:obj_eval']}.
  • Figure 3: The predicted waveforms of a short segment from one of the m1 test samples. The differences were computed by subtracting the predicted signal from the reference.
  • Figure 4: The MOS results of the vocoders trained on different singers with 95% confidence interval.
  • Figure :