Year: 2026
Type: preprint
Abstract: 5,992 bona fide speech clips (VCTK and AMI) passed through 34 speech and audio codecs, 12 tandem codec chains and 5 resample-only controls. Every condition holds the same clips, and every clip has a per-clip record of the exact commands that produced it. No synthetic or spoofed speech. See README.md in this record before measuring: sample rate and bandwidth are confounds, and the resample controls are there to separate them from codec effects.
Source: arXiv (Cornell University)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7084446708
Cites:
Cited by:
Related to: