Skip to content
SPECIFICTION
Get listed

Pella Research

pellaresearch.com

WebAPI
Visit pellaresearch.com
Spec score 7.2 #16 of 20 in AI Deepfake Detection Tools

What it is

Pella Research studies synthetic speech and builds models for detecting it in audio. Its research examines vocoder artifacts, spectral statistics, and prosodic cues. A browser demo accepts an uploaded or recorded voice clip and returns a verdict without signup; the site says audio is not stored. The REST detection API returns calibrated authenticity scores and per-segment analysis, with access reviewed after a request, typically within 24 hours. Pella offers the API to journalists, platforms, researchers, and institutions, and lists a free tier for accredited journalists and academics. Published evaluation sets track new voice-cloning systems with documented provenance and no train/test leakage. Pella says models are trained and tested with phone-line compression, background noise, and re-encoding. It reports pellav2 results of 95.82% accuracy, 0.959 F1, and 57 ms latency in the August 2026 Podonos benchmark, plus 93.6% accuracy on m4a files with errors leaning toward false positives. pellav2 is described as open weights under the MIT license.

Spec sheet

Purpose
Pella Research studies synthetic speech and builds models to detect it in real-world audio.Read 7 Oct 2026 · pellaresearch.com
Product
Pella Research builds models to detect synthetic speech by analyzing vocoder artifacts, spectral statistics, and prosodic cues.Read 7 Oct 2026 · pellaresearch.com
Intended users
Pella describes its work as serving platforms, journalists, courts, and researchers investigating whether audio is synthetic.Read 7 Oct 2026 · pellaresearch.com
API
The detection API is a REST endpoint that returns calibrated authenticity scores and per-segment analysis, with access granted on request.Read 7 Oct 2026 · pellaresearch.com
License
The maker lists pellav2 as open weights under the MIT license.Read 7 Oct 2026 · pellaresearch.com
Headquarters
Amsterdam, NetherlandsRead 28 Sep 2026 · pellaresearch.com
Access and support
API access is reviewed on request for journalists, platforms, researchers, and institutions, with a typical response within 24 hours.Read 7 Oct 2026 · pellaresearch.com
Access response
Pella says it reviews API requests and typically responds within 24 hours.Read 7 Oct 2026 · pellaresearch.com
API audience
The API is offered to journalists, platforms, researchers, and institutions, and the page lists a free tier for accredited journalists and academics.Read 7 Oct 2026 · pellaresearch.com
Benchmark limitation
Pella reports 93.6% accuracy on m4a files and says its errors lean toward false positives.Read 7 Oct 2026 · pellaresearch.com
Benchmark result
Pella reports that pellav2 achieved 95.82% accuracy, a 0.959 F1 score, and 57 ms latency in the August 2026 Podonos benchmark.Read 7 Oct 2026 · pellaresearch.com
Benchmarks
Pella says its evaluation sets track new voice cloning systems and use documented provenance with no train-test leakage.Read 7 Oct 2026 · pellaresearch.com
Demo limitation
The live demo runs on free shared hardware, and its first request after a quiet period can take up to a minute to wake up.Read 7 Oct 2026 · pellaresearch.com
Free access
Pella states that accredited journalists and academics can receive a free API tier.Read 7 Oct 2026 · pellaresearch.com
Known failure case
Pella reports 93.6% accuracy on m4a files in that benchmark and says its errors lean toward false positives.Read 7 Oct 2026 · pellaresearch.com
Live demo
The browser demo lets users upload or record a voice clip and receive a verdict without signup; the page says no audio is stored.Read 7 Oct 2026 · pellaresearch.com
Location
The site lists Amsterdam, Netherlands.Read 7 Oct 2026 · pellaresearch.com
Model availability
The site describes pellav2 as open weights under the MIT license and says the public checkpoint reproduces the submitted benchmark score.Read 7 Oct 2026 · pellaresearch.com
Real-world conditions
Pella says its models are trained and tested with phone-line compression, background noise, and re-encoding.Read 7 Oct 2026 · pellaresearch.com
Research
Its research examines vocoder artifacts, spectral statistics, and prosodic cues left by speech synthesis.Read 7 Oct 2026 · pellaresearch.com
Research and benchmarks
It publishes evaluation sets that track new voice cloning systems, with documented provenance and no train/test leakage.Read 7 Oct 2026 · pellaresearch.com
Robustness
Pella says its models are trained and tested on audio affected by phone-line compression, background noise, and re-encoding.Read 7 Oct 2026 · pellaresearch.com

Strong points and limits

Strong points

  • No-signup browser voice demo
  • Per-segment API analysis
  • Free API tier for accredited journalists and academics
  • Evaluation sets with documented provenance
  • Open weights under MIT license

Limits

  • API access requires review
  • Demo wake-up can take up to a minute
  • 93.6% reported accuracy on m4a files
  • Errors reportedly lean toward false positives

Who it suits

It is intended for platforms, journalists, courts, researchers, and institutions investigating whether audio is synthetic. The browser demo supports quick checks, while API access supports programmatic analysis.

Verdict

Pella offers both an immediate browser check and an API for calibrated synthetic-speech analysis. Review its reported m4a limitation and request API access for organizational use.

Where Specifiction places it

Alternatives to Pella ResearchAll alternatives →