Pella Research

What Pella Research is

Pella Research builds models to assess whether audio is synthetic by examining vocoder artifacts, spectral statistics, and prosodic cues. Its browser demo lets users upload or record a voice clip and receive a verdict without signing up; the site says audio is not stored. A REST API returns calibrated authenticity scores and per-segment analysis, with access reviewed on request. Pella says it typically responds to requests within 24 hours, and accredited journalists and academics can receive a free API tier. The organization also publishes evaluation sets that track new voice-cloning systems, with documented provenance and no train/test leakage. Pella says its models are trained and tested on audio affected by phone-line compression, background noise, and re-encoding. The site reports that pellav2 achieved 95.82% accuracy, a 0.959 F1 score, and 57 ms latency in the August 2026 Podonos benchmark. It reports 93.6% accuracy on m4a files and says errors lean toward false positives. The maker lists pellav2 as open weights under the MIT license.