Audio Examples
Audio examples for a submission to ICASSP 2026: Generative Modeling of Text-Aligned Speech Tokens via Discrete Diffusion
All audio examples are randomly selected from the librispeech test-clean set.
Comparison between auto-regressive and discrete-diffusion models
For DDMs, we use the ancestral-conf-top-k sampler with 50 inference steps.
8455-210777-0066
original
auto-regressive
discrete-diffusion
8555-284447-0022
original
auto-regressive
discrete-diffusion
4992-41806-0005
original
auto-regressive
discrete-diffusion
5639-40744-0019
original
auto-regressive
discrete-diffusion
8224-274381-0002
original
auto-regressive
discrete-diffusion
