Adversarial Audio Synthesis
lastmod 2026-11-10created 2020-01-092 edits
Adversarial Audio Synthesis is a paper that proposed generating raw audio waveforms directly with generative adversarial networks (GANs). It compares WaveGAN, which adapts the DCGAN architecture for image generation to one-dimensional waveforms, with SpecGAN, which generates spectrograms like images and then converts them back to audio, and shows experiments synthesizing short sounds such as spoken digits, drums, and bird calls. Unlike autoregressive models, it produces the whole waveform at once, so generation is fast.
Adversarial Audio Synthesis
Documentation
- Adversarial Audio Synthesis
- https://arxiv.org/abs/1802.04208