Picovoice/porcupine

On-device wake word detection powered by deep learning

What it solves

Porcupine is a lightweight, highly accurate wake word engine designed to enable "always-listening" voice-enabled applications. It allows a system to efficiently detect specific, static voice commands (wake words) in real-time without consuming significant computational resources.

How it works

The engine uses deep neural networks trained in real-world environments to identify specific audio patterns. It is designed to be compact and computationally efficient, making it suitable for deployment on everything from high-end desktops to low-power IoT devices and microcontrollers. Developers can use pre-built keywords or train custom wake word models via the Picovoice Console.

Who it’s for

Developers building voice-activated interfaces for IoT devices, mobile apps (Android/iOS), web applications, and embedded systems (like Raspberry Pi or Arduino) who need a reliable way to to trigger actions based on a few static voice commands.

Highlights

  • Extensive Cross-Platform Support: Works across Arm Cortex-M, STM32, Arduino, Raspberry Pi, Android, iOS, Web (Chrome, Safari, Firefox, Edge), Linux, macOS, and Windows.
  • High Performance: Claimed to be 11 times more accurate and 6.5 times faster than competing libraries like PocketSphinx and Snowboy on Raspberry Pi 3.
  • Customizable: Supports custom wake word training through a self-service console.
  • Multilingual: Supports English, Chinese (Mandarin), French, German, Italian, Japanese, Korean, Portuguese, and Spanish.
  • Scalable: Can detect multiple wake words simultaneously without increasing the runtime footprint.

Related

  • Project
    wenet-e2e/wekws

    A production-ready end-to-end toolkit for small-footprint wake-up word detection designed for low-power IoT devices.

  • Project
    cmusphinx/pocketsphinx

    PocketSphinx is an open‑source, offline speech‑to‑text engine (C library with Python bindings) that runs efficiently on low‑resource devices. It provides a command‑line tool and APIs for recognizing or aligning audio, outputting results as JSON.

  • Project
    Picovoice/rhino

    Rhino is Picovoice’s on‑device speech‑to‑intent engine. It runs a compact neural network on any platform (micro‑controllers, phones, browsers, desktop) and returns a structured intent plus slot values from spoken commands. Developers define custom command sets (contexts) via the Picovoice Console, then use thin SDKs (Python, .NET, Java, C, Flutter, React‑Native, Android, iOS, Web, Node) to process 16‑bit PCM audio in real time. Ideal for low‑latency, privacy‑preserving voice control in IoT, mobile, and web apps.

  • Project
    jxlarrea/voice-satellite-card-integration

    A Home Assistant integration that turns any browser or tablet into a private voice assistant satellite with on-device wake word detection and visual feedback.

  • Project
    collabora/WhisperLive

    WhisperLive is an open‑source, near‑real‑time transcription server built around OpenAI’s Whisper model. It supports multiple inference back‑ends (faster‑whisper, NVIDIA TensorRT, Intel OpenVINO, AMD ROCm), offers word‑level timestamps, hot‑word boosting, speaker diarization and optional translation, and provides both simple command‑line clients and a Python streaming API. Docker images are provided for GPU and CPU deployments, and browser/iOS extensions let you transcribe audio directly from a web page or mobile device.