We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Sound

Title: DDSP-based Singing Vocoders: A New Subtractive-based Synthesizer and A Comprehensive Evaluation

Abstract: A vocoder is a conditional audio generation model that converts acoustic features such as mel-spectrograms into waveforms. Taking inspiration from Differentiable Digital Signal Processing (DDSP), we propose a new vocoder named SawSing for singing voices. SawSing synthesizes the harmonic part of singing voices by filtering a sawtooth source signal with a linear time-variant finite impulse response filter whose coefficients are estimated from the input mel-spectrogram by a neural network. As this approach enforces phase continuity, SawSing can generate singing voices without the phase-discontinuity glitch of many existing vocoders. Moreover, the source-filter assumption provides an inductive bias that allows SawSing to be trained on a small amount of data. Our experiments show that SawSing converges much faster and outperforms state-of-the-art generative adversarial network and diffusion-based vocoders in a resource-limited scenario with only 3 training recordings and a 3-hour training time.
Comments: Accepted at ISMIR 2022
Subjects: Sound (cs.SD); Audio and Speech Processing (eess.AS)
Journal reference: International Society for Music Information Retrieval (ISMIR) 2022
Cite as: arXiv:2208.04756 [cs.SD]
  (or arXiv:2208.04756v2 [cs.SD] for this version)

Submission history

From: Wen-Yi Hsiao [view email]
[v1] Tue, 9 Aug 2022 13:06:08 GMT (6406kb,D)
[v2] Fri, 19 Aug 2022 03:19:23 GMT (6406kb,D)

Link back to: arXiv, form interface, contact.