Cobra VAD
The accuracy-first on-device VAD: Picovoice's engine that reports 98.9% true-positive rate at 5% false-positive rate with ~5% CPU on a Raspberry Pi Zero — fully local, no audio ever leaves the device.
About
Cobra is Picovoice's voice activity detection engine, built for the accuracy end of the VAD trade-off. Where classic detectors like WebRTC VAD are energy/GMM-based and leak anything loud and voice-like (TVs, café chatter, music with vocals), Cobra is a neural engine tuned for hard negatives. Picovoice backs this with an open benchmark (Apache 2.0 code, LibriSpeech + DEMAND noise, 18 environments) publishing ROC curves against WebRTC and Silero; their docs show a real-time factor of 0.0002 on a laptop CPU and 0.037 on a Raspberry Pi Zero. An independent 2026 comparison table puts Cobra at 98.9% TPR at 5% FPR — the best of the three compared. The trade: it's proprietary. The SDKs (Python, Node, .NET, web, iOS, Android, Raspberry Pi) are Apache 2.0, but the engine needs a free AccessKey from the Picovoice Console and production use falls under Picovoice's commercial terms. Cross-platform coverage is excellent — it runs in the browser via WebAssembly too. Choose it when compliance or accuracy demands the best on-device VAD and you can live with a vendor key; choose Silero when MIT purity matters more than the last few accuracy points.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- 98.9% TPR at 5% FPR (third-party 2026 comparison table; WebRTC VAD 50.0%, Silero 87.7%)\n- RTF 0.0002 on laptop CPU (Intel Core Ultra 7), 0.037 on Raspberry Pi Zero (provider-reported, open benchmark)\n- Provider publishes full ROC curves on LibriSpeech + DEMAND at 0dB SNR; no independent peer-reviewed benchmark found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: actively maintained engine (repo pushed October 2026), open Apache 2.0 benchmark with published ROC curves, top accuracy in third-party 2026 comparison. Strengths: best-in-class accuracy among tested VADs, genuinely cross-platform including WASM, fully on-device. Docked for: (1) proprietary engine — weights not inspectable, commercial terms for production; (2) headline accuracy figures are provider-measured on a provider-built benchmark; (3) requires a vendor AccessKey. License: SDKs Apache 2.0; engine proprietary (free AccessKey for dev).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 4.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · Top accuracy among compared VADs: 98.9% TPR at 5% FPR in a 2026 third-party comparison table.
- · Open Apache 2.0 benchmark (LibriSpeech + DEMAND) with published ROC curves.
- · Fully on-device including WASM; requires vendor AccessKey.