Cobra
Voice Activity Detection

powered by deep learning, efficient, and hyper cross-platform

Start Building
Talk to Sales

Accurate

Cobra outperforms existing solutions with wide margins. Compared to WebRTC VAD, Cobra achieves a much higher true-positive rate at any given false-positive operating point. Picovoice is the only speech technology provider that provides open-source and reproducible benchmarks. [1]

Leopard Speech-to-Text accuracy compared to IBM ASR, Google speech recognition, AWS transcribe, and Azure STT

Private

All voice data is processed on-device. Intrinsically HIPAA and GDPR compliant.

Zero-Latency

Edge-first architecture eliminates unpredictable network delay.

Free Tier

Start building for free. Small projects can remain free, forever.

Try it!
Arrow (pointing at microphone button)
Press the microphone button to activate the demo.
cobra = pvcobra.create('${ACCESS_KEY}')
while True:
is_voiced = cobra.process(audio_frame())
Build with Python
Cobra cobra = new Cobra("${ACCESS_KEY}");
while(true) {
float isVoiced = cobra.process(audioFrame());
}
Build with Android
let cobra = Cobra(accessKey: "${ACCESS_KEY}")
while true {
let isVoiced = cobra.process(audioFrame())
}
Build with iOS
let worker = await CobraWorkerFactory.create(
"${ACCESS_KEY}",
(isVoiced) => {});
const processor = await WebVoiceProcessor.init(
{engines: [worker]});
processor.start();
Build with Web