Format path
Six common browser formats have completed live normalization and API checks with HTTP 200.
AI voice and speech signal
Route audio.
Abstain safely.
MP3, M4A, AAC, OGG, FLAC and WAV are normalized locally to 16 kHz mono PCM. Speech can receive a weak authenticity signal; music, mixed and other audio return unknown.
Open the free checkerScope before score.
Every result is a warning signal. The evidence and limitations below define what this route can and cannot support.
Speech evidence
The frozen 193-clip source-disjoint speech holdout reached 84.46% balanced accuracy and 15.53% real false positives.
Semantic router
YAMNet separates speech, music, mixed and other content. It is not itself an authenticity detector.
Do not infer more than the lane measured.
- A voice score does not identify the speaker or prove authenticity.
- Music, mixed, tonal, quiet and unsupported clips can return unknown.
- General audio detection remains experimental and unsuitable for consequential automation.