Mapping the physical reality of human and synthetic voices.
What we do
We build deterministic acoustic measurement and the reference datasets that go with it. Our methods come from classical physics, digital signal processing and statistics. Results are reported in units a person can check by hand, and every input and every step is on the record, so someone who has never spoken to us can reproduce them.
How we work
Deterministic math
Same input, same output. Fixed seeds, pinned software versions and classical methods. We do not guess; we measure.
Human-interpretable units
Decibels, hertz, seconds, semitones and kilobits per second. Conditions and features are named for what was physically done or measured.
Provenance and reproducibility
Every clip traces to a named source and licence. Files carry SHA-256 hashes, and per-clip records log the seeds and commands that produced them.
Datasets
PACC-T
Telecoms. The same 5,992 bona fide speech clips passed through 34 codecs, 12 tandem chains and 5 resample controls.
CC BY 4.0 · v1.0
PACC-P
Presentation. The same 5,992 bona fide speech clips under 50 presentation conditions: noise, babble, music, rooms, echo, band limiting, pitch and tempo changes.
CC BY 4.0 · v1.0
Both corpora hold bona fide speech only, with no synthetic or spoofed speech. About the corpus
Provenance
Where every clip came from, under which licence, and how to verify what you download.
Methods & Provenance Reference Library