Porcupine

// Overview

Porcupine is a wake word engine from Picovoice for building always-listening, voice-enabled applications. It solves one narrowly scoped problem well: detecting one or a few static trigger phrases — the "Hey Siri" moment — so your application knows when to start paying attention. The models are trained on real-world audio and the engine is described as compact and computationally efficient, which is what makes it viable on an IoT device rather than only on a workstation.

Its defining property is that detection happens entirely on-device. Nothing is streamed to a server to decide whether the trigger phrase was spoken, which matters both for privacy and for latency. The engine is also scalable in an unusual sense: it can listen for multiple wake words at once with no added runtime footprint, so a device can respond to several distinct commands without running several detectors.

Porcupine is deliberately not a general speech-recognition system. Picovoice points you at its own Rhino Speech-to-Intent engine when you need to understand complex, naturally spoken commands within a domain, and at the broader Picovoice platform when you want an Alexa-style end-to-end voice experience. The repository code is Apache-2.0, but the SDKs authenticate with an AccessKey obtained from Picovoice Console, which is also where you train custom wake word models.

// What it does

• Runs fully on-device — no audio leaves the machine for the wake-word decision
• Cross-platform by design: Arm Cortex-M, STM32 and Arduino, Raspberry Pi, Android and iOS, the major browsers, and Linux, macOS and Windows
• SDKs for Python, .NET, Java, C, Node.js, Flutter, React Native and the web, published to the usual package registries
• Detects multiple always-listening commands at once with no added runtime footprint
• Ships a set of default keyword models, listable at runtime, plus custom wake words trained self-service in Picovoice Console
• Nine supported languages including English, Mandarin, French, German, Italian, Japanese, Korean, Portuguese and Spanish

// Getting started

Every SDK follows the same shape: install the package for your platform, supply an AccessKey from Picovoice Console, then create an engine instance for the keywords you want to listen for. The Python SDK is the shortest path to a first run.

pip3 install pvporcupine

import pvporcupine

# AccessKey obtained from Picovoice Console (https://console.picovoice.ai/)
access_key = "${ACCESS_KEY}"

handle = pvporcupine.create(access_key=access_key, keywords=['picovoice', 'bumblebee'])

import pvporcupine

print(pvporcupine.KEYWORDS)

sudo pip3 install pvporcupinedemo

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Add a hands-free trigger phrase to a mobile or desktop app so it starts listening without a tap
• Run always-listening voice activation on a Raspberry Pi or a Cortex-M microcontroller where a cloud round trip is not an option
• Give a browser-based assistant a wake word without streaming continuous microphone audio to a server
• Listen for several distinct spoken commands on one device without paying for several detectors

// How Porcupine compares

Porcupine alongside other open-source audio, music & voice tools AI/TLDR tracks, ranked by GitHub stars.