On-device wake-word detection engine powered by deep learning.

Porcupine

Made in Vancouver, Canada by Picovoice

Porcupine is a highly-accurate and lightweight wake word engine. It enables building always-listening voice-enabled applications. It is

• using deep neural networks trained in real-world environments.
• compact and computationally-efficient. It is perfect for IoT.
• cross-platform. Raspberry Pi, BeagleBone, Android, iOS, Linux (x86_64), macOS (x86_64), Windows (x86_64), and web browsers are supported. Additionally, enterprise customers have access to the ARM Cortex-M SDK.
• scalable. It can detect multiple always-listening voice commands with no added runtime footprint.
• self-service. Developers can train custom wake word models using Picovoice Console.

Porcupine wake word models for all major voice assistants (Alexa, Hey Google, Ok Google, and Hey Siri) are available for free (under Apache 2.0) in this repo.

Table of Contents

• Porcupine
  • Table of Contents
  • License & Terms
  • Use Cases
  • Try It Out
  • Performance
  • Demos
  • Python
  • .NET
  • Java
  • React Native
  • Android
  • iOS
  • JavaScript
  • NodeJS
  • C
  • SDKs
  • Python
  • .NET
  • Java
  • React Native
  • Android
  • iOS
  • JavaScript
  • NodeJS
  • C
  • Releases
  • FAQ

License & Terms

Porcupine SDK is free and licensed under Apache 2.0 including the models released within the repository. Picovoice Console offers two types of subscriptions: Personal and Enterprise. Personal accounts can train custom wake word models that run on x86_64, subject to limitations and strictly for non-commercial purposes. Personal accounts empower researchers, hobbyists, and tinkerers to experiment. Enterprise accounts can unlock all capabilities of Picovoice Console, are permitted for use in commercial settings, and have a path to graduate to commercial distribution*.

Use Cases

Porcupine is the right product if you need to detect one or a few static (always-listening) voice commands.

• If you want to create voice experiences similar to Alexa or Google, see the Picovoice platform.
• If you need to understand complex and naturally-spoken voice commands within a specific domain, see the Rhino Speech-to-Intent engine.

Try It Out

• Interactive Web Demo

• Android Demo Application

• Porcupine and Rhino on an ARM Cortex-M7

Performance

A comparison between accuracy and runtime metrics of Porcupine and two other widely-used libraries, PocketSphinx and Snowboy, is provided here. Compared to the best-performing engine of these two, Porcupine is 6.0 times more accurate and 6.5 times faster (on Raspberry Pi 3).

Demos

Python Demos

Install PyAudio and then the demo package:

sudo pip3 install pvporcupinedemo

With a working microphone connected to your device run the following in the terminal:

porcupine_demo_mic --keywords porcupine

The engine starts processing the audio input from the microphone in realtime and outputs to the terminal when it detects utterances of Porcupine.

For more information about Python demos go to demo/python.

.NET Demos

Install OpenAL and then from demo/dotnet/PorcupineDemo run the following in the terminal to build the demo:

dotnet build -c MicDemo.Release

Make sure there is a working microphone connected to your device. From demo/dotnet/PorcupineDemo run the following in the terminal:

dotnet run -c MicDemo.Release -- --keywords porcupine

The engine starts processing the audio input from the microphone in realtime and outputs to the terminal when it detects utterances of Porcupine.

For more information about .NET demos go to demo/dotnet.

Java Demos

Make sure there is a working microphone connected to your device. From the root of the repository, run the following command from the terminal:

java -jar demo/java/bin/porcupine-mic-demo.jar -k porcupine

The engine starts processing the audio input from the microphone in realtime and outputs to the terminal when it detects utterances of Porcupine.

For more information about Java demos go to demo/java.

React Native Demos

To run the React Native Porcupine demo app you will first need to setup your React Native environment. For this, please refer to React Native's documentation. Once your environment has been set up, navigate to demo/react-native to run the following commands:

For Android:

yarn android-install    # sets up environment
yarn android-run        # builds and deploys to Android

For iOS:

yarn ios-install        # sets up environment
yarn ios-run            # builds and deploys to iOS

Android Demos

Using Android Studio, open demo/android/Activity as an Android project and then run the application.

To learn about how to use Porcupine in long running services go to demo/android/Service.

iOS Demos

Using Xcode, open PorcupineDemo and run the application.

JavaScript Demos

From demo/javascript run the following in the terminal:

yarn
yarn start

If you prefer to use NPM instead, use the following:

npm install
npm install -g copy-files-from-to
copy-files-from-to
npx serve

This will launch a local server running the demo. Open http://localhost:5000 in your web browser and follow the instructions on the page.

NodeJS Demos

Install node-record-lpcm16 NPM package and follow the instructions there for setting up your microphone. Then install the demo package:

yarn global add @picovoice/porcupine-node-demo

With a working microphone connected to your device run the following in the terminal:

ppn-mic-demo --keywords porcupine

The engine starts processing the audio input from the microphone in realtime and outputs to the terminal when it detects utterances of Porcupine.

For more information about NodeJS demos go to demo/nodejs.

C Demos

Microphone demo runs on Linux-based systems (e.g. Ubuntu, Raspberry Pi, and BeagleBone). Build the demo:

gcc -std=c99 -O3 -o demo/c/porcupine_demo_mic \
-I include/ demo/c/porcupine_demo_mic.c -ldl -lasound

Find the name of audio input device (microphone) on your computer using arecord -L and then from the root of the repository run the demo:

./demo/c/porcupine_demo_mic ${LIBRARY_PATH} lib/common/porcupine_params.pv \
resources/keyword_files/${SYSTEM}/porcupine_${SYSTEM}.ppn 0.5 ${INPUT_AUDIO_DEVICE}

Replace ${LIBRARY_PATH} with path to appropriate library available under lib, ${SYSTEM} with the name of the platform you are running on (linux, raspberry-pi, or beaglebone), and ${INPUT_AUDIO_DEVICE} with the name of your microphone device. The demo opens an audio stream and detects utterances of Porcupine.

For more information about C demos go to demo/c.

SDKs

Python

Install the Python SDK:

pip3 install pvporcupine

The SDK exposes a factory method to create instances of the engine:

import pvporcupine

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

keywords argument is a shorthand for accessing default keyword files shipped with the library. The default keyword files available can be retrieved via

import pvporcupine

print(pvporcupine.KEYWORDS)

If you wish to use a non-default keyword file you need to identify its path:

import pvporcupine

handle = pvporcupine.create(keyword_paths=['path/to/non/default/keyword/file'])

When initialized, valid sample rate can be obtained using handle.sample_rate. The required frame length (number of audio samples in an input array) is handle.frame_length. The object can be used to monitor incoming audio as follows:

import pvporcupine

handle = pvporcupine.create(keywords=['porcupine'])

def get_next_audio_frame():
    pass

while True:
    keyword_index = handle.process(get_next_audio_frame())
    if keyword_index >= 0:
        # Insert detection event callback here
        pass

Finally, when done be sure to explicitly release the resources using handle.delete().

.NET

Install the .NET SDK using NuGet or the dotnet CLI:

dotnet add package Porcupine

The SDK exposes a factory method to create instances of the engine:

using Pv

Porcupine handle = Porcupine.Create(keywords: new List<string> { "picovoice" });

The keywords argument is a shorthand for accessing built-in keyword files shipped with the library. The built-in keyword files available can be retrieved via:

using Pv

foreach (string keyword in Porcupine.KEYWORDS)
{
    Console.WriteLine(keyword);
}

If you wish to use a custom keyword file (i.e. a keyword file generated by Picovoice Console, with a .ppn extension), you need to specify its path:

using Pv

Porcupine handle = Porcupine.Create(
    keywordPaths: new List<string>{ "path/to/custom/keyword/file"});

When initialized, the required sample rate can be obtained using handle.SampleRate. Expected frame length (number of audio samples in an input array) is handle.FrameLength. The object can be used to monitor incoming audio as below:

short[] getNextAudioFrame()
{
    // .. get a frame of audio
    return audioFrame;
}

while(true)
{
    var keywordIndex = handle.Process(getNextAudioFrame())
    if(keywordIndex >= 0)
    {
        // .. Insert detection event callback here
    }
}

Porcupine will have its resources freed by the garbage collector, but to have resources freed immediately after use, wrap it in a using statement:

using(Porcupine handle = Porcupine.Create(keywords: new List<string> { "picovoice" }))
{
    // .. Porcupine usage here
}

Java

Install the Porcupine Java binding by downloading and referencing the latest Porcupine JAR file available here. The SDK exposes a builder to create instances of the engine:

import ai.picovoice.porcupine.*;

try{
    Porcupine handle = new Porcupine.Builder()
                        .setKeyword("picovoice")
                        .build();
} catch (PorcupineException e) { }

The setKeyword() builder argument is a shorthand for accessing built-in keyword model files shipped with the package.

The built-in keyword files available can be retrieved via:

import ai.picovoice.porcupine.*;

for(String keyword : Porcupine.KEYWORDS){
    System.out.println(keyword);
}

If you wish to use a custom keyword file (i.e. a keyword file generated by Picovoice Console, with a .ppn extension) you need to the file path as demonstrated below:

import ai.picovoice.porcupine.*;

try{
    Porcupine handle = new Porcupine.Builder()
                        .setKeywordPath("path/to/custom/keyword/file")
                        .build();
} catch (PorcupineException e) { }

When initialized, valid sample rate can be obtained using handle.getSampleRate(). Expected frame length (number of audio samples in an input array) is handle.getFrameLength(). The object can be used to monitor incoming audio as below:

short[] getNextAudioFrame(){
    // .. get audioFrame
    return audioFrame;
}

while(true){
    int keywordIndex = handle.Process(getNextAudioFrame());
    if(keywordIndex >= 0){
        // .. detection event logic/callback
    }
}

Once you're done with Porcupine, ensure you release its resources explicitly:

handle.delete();

React Native

Install @picovoice/react-native-voice-processor and @picovoice/porcupine-react-native. The SDK provides two APIs:

High-Level API

PorcupineManager provides a high-level API that takes care of audio recording. This class is the quickest way to get started.

Using the constructor PorcupineManager.fromKeywords will create an instance of the PorcupineManager using one or more of the built-in keywords.

async createPorcupineManager(){
    try{
        this._porcupineManager = await PorcupineManager.fromKeywords(
            ["picovoice", "porcupine"],
            detectionCallback);
    } catch (err) {
        // handle error
    }
}

To create an instance of PorcupineManager that detects custom keywords, you can use the PorcupineManager.fromKeywordPaths static constructor and provide the paths to the .ppn file(s).

this._porcupineManager = await PorcupineManager.fromKeywords(
  ["/path/to/keyword.ppn"],
  detectionCallback
);

Once you have instantiated a Porcupine manager, you can start/stop audio capture and wake word detection by calling:

let didStart = this._porcupineManager.start();
// .. use Porcupine
let didStop = this._porcupineManager.stop();

Once the app is done with using PorcupineManager, be sure you explicitly release the resources allocated to Porcupine:

this._porcupineManager.delete();

There is no need to deal with audio capture to enable wake word detection with PorcupineManager. This is because it uses @picovoice/react-native-voice-processor module to capture frames of audio and automatically pass it to the wake word engine.

Low-Level API

Porcupine provides low-level access to the wake word engine for those who want to incorporate wake word detection into a already existing audio processing pipeline. Porcupine also has fromKeywords and fromKeywordPaths static constructors.

async createPorcupine(){
    try{
        this._porcupine = await Porcupine.fromKeywords(["picovoice"]);
    } catch (err) {
        // handle error
    }
}

In this case you do not pass in a detection callback as you will be passing in audio frames directly using the process function:

let buffer = getAudioFrame();

try {
  let keywordIndex = await this._porcupine.process(buffer);
  if (keywordIndex >= 0) {
    // detection made!
  }
} catch (e) {
  // handle error
}

For process to work correctly, the provided audio must be single-channel and 16-bit linearly-encoded.

Finally, once you no longer need the wake word engine, be sure to explicitly release the resources allocated to Porcupine:

this._porcupine.delete();

Android

There are two possibilities for integrating Porcupine into an Android application.

High-Level API

[PorcupineManager](binding/android/Porcupine/porcupine/src/main/java/ai/picovoice/porcupine/PorcupineManager.java) provides a high-level API for integrating Porcupine into Android applications. It manages all activities related to creating an input audio stream, feeding it into the Porcupine library, and invoking a user-provided detection callback.

final String modelPath = ... // Available at lib/common/porcupine_params.pv
final String keywordPath = ...
final float sensitivity = 0.5f;

PorcupineManager manager = new PorcupineManager(
        modelPath,
        keywordPath,
        sensitivity,
        new PorcupineManagerCallback() {
            @Override
            public void invoke(int keywordIndex) {
                // detection event logic/callback
            }
        });

Sensitivity is the parameter that enables developers to trade miss rate for false alarm. It is a floating point number within [0, 1]. A higher sensitivity reduces miss rate at cost of increased false alarm rate.

When initialized, input audio can be monitored using manager.start(). Stop the manager using by invoking manager.stop(). When done be sure to release the resources using manager.delete().

Low-Level API

Porcupine provides a binding for Android. It can be initialized using.

final String modelPath = ... // Available at lib/common/porcupine_params.pv
final String keywordPath = ...
final float sensitivity = 0.5f;

Porcupine porcupine = new Porcupine(modelPath, keywordPath, sensitivity);

Sensitivity is the parameter that enables developers to trade miss rate for false alarm. It is a floating point number within [0, 1]. A higher sensitivity reduces miss rate at cost of increased false alarm rate.

Once initialized, porcupine can be used to monitor incoming audio.

private short[] getNextAudioFrame();

while (true) {
    final int keywordIndex = porcupine.process(getNextAudioFrame());
    if (keywordIndex != -1) {
        // detection event logic/callback
    }
}

Finally, be sure to explicitly release resources acquired by porcupine as the binding class does not rely on the garbage collector for releasing native resources.

porcupine.delete();

iOS

There are two approaches for integrating Porcupine into an iOS application.

High-Level API

PorcupineManager manages audio recording, passing it into Porcupine, and invoking the user-provided detection callback.

let modelPath: String = ... // Available at lib/common/porcupine_params.pv
let keywordPaths: [String] = ["/path/to/keyword/file/a", "/path/to/keyword/file/b"]
let sensitivities: [Float32] = [0.35, 0.64]
let keywordCallback: ((Int32) -> Void) = { keywordIndex in
    // Insert detection event logic
}

let manager = try PorcupineManager(
    modelPath: modelPath,
    keywordPaths: keywordPaths,
    sensitivities: sensitivities
    onDetection: keywordCallback)

When initialized, input audio can be monitored using manager.start(). When done be sure to stop the manager using manager.stop().

Direct

Porcupine is shipped as a precompiled ANSI C library and can directly be used in Swift using module maps. It can be initialized to detect multiple wake words concurrently using:

let modelPath: String = ... // Available at lib/common/porcupine_params.pv
let keywordPaths: [String] = ["/path/to/keyword/file/a", "/path/to/keyword/file/b"]
let sensitivities: [Float32] = [0.35, 0.64]

var handle: OpaquePointer?
let status = pv_porcupine_init(
    modelPath,
    Int32(keywordFilePaths.count),
    keywordPaths.map{ UnsafePointer(strdup($0)) },
    sensitivities,
    &handle)
if status != PV_STATUS_SUCCESS {
    // error handling logic
}

Then handle can be used to monitor incoming audio stream.

func getNextAudioFrame() -> UnsafeMutablePointer<Int16> {
    //
}

while true {
    let pcm = getNextAudioFrame()
    var keyword_index: Int32 = -1

    let status = pv_porcupine_process(handle, pcm, &keyword_index)
    if status != PV_STATUS_SUCCESS {
        // error handling logic
    }
    if keyword_index >= 0 {
        // detection event logic/callback
    }
}

When finished, release the resources via

pv_porcupine_delete(handle)

JavaScript

Porcupine is available on modern web browsers via WebAssembly. The JavaScript binding makes it easy to use Porcupine within a browser environment.

Instantiate a new instance of engine using the factory method:

let keywordModels = [new Uint8Array([...]), ...];
let sensitivities = new Float32Array([0.5, ...]);

let handle = Porcupine.create(keywordModels, sensitivities)

When instantiated, handle can process audio via its .process method.

let getNextAudioFrame = function() {
    ...
};

while (true) {
    let keywordIndex = handle.process(getNextAudioFrame());
    if (keywordIndex !== -1) {
        // Insert detection callback
    }
}

When done be sure to release resources acquired by WebAssembly using release():

handle.release();

NodeJS

Install NodeJS SDK:

yarn add @picovoice/porcupine-node

Create instances of the Porcupine class by specifying which keywords you want it to listen for:

const Porcupine = require("@picovoice/porcupine-node");

const {
  GRASSHOPPER,
  BUMBLEBEE,
} = require("@picovoice/porcupine-node/builtin_keywords");

let handle = new Porcupine([GRASSHOPPER, BUMBLEBEE], [0.5, 0.65]);

GRASSHOPPER and BUMBLEBEE are built-in keywords. If you wish to use a custom keyword file, you need to identify its path:

const Porcupine = require("@picovoice/porcupine-node");

let handle = new Porcupine(["/path/to/custom/keyword/file"], [0.5]);

When instantiated, handle can process audio via its .process method.

let getNextAudioFrame = function() {
    ...
};

while (true) {
  let keywordIndex = handle.process(getNextAudioFrame());
  if (keywordIndex !== -1) {
    // detection event callback
  }
}

When done be sure to release resources acquired by WebAssembly using release():

handle.release();

C

Porcupine is implemented in ANSI C and therefore can be directly linked to C applications. include/pv_porcupine.h header file contains relevant information. An instance of Porcupine object can be constructed as follows.

const char *model_path = ... // Available at lib/common/porcupine_params.pv
const char *keyword_path = ...
const float sensitivity = 0.5f;

pv_porcupine_t *handle = NULL;
const pv_status_t status = pv_porcupine_init(
    model_path,
    1,
    &keyword_path,
    &sensitivity,
    &handle);
if (status != PV_STATUS_SUCCESS) {
    // Insert error handling logic
}

Sensitivity is the parameter that enables developers to trade miss rate for false alarm. It is a floating point number within [0, 1]. A higher sensitivity reduces miss rate (false reject rate) at cost of (potentially) increased false alarm rate.

Now the handle can be used to monitor incoming audio stream. Porcupine accepts single channel, 16-bit linearly-encoded PCM audio. The sample rate can be retrieved using pv_sample_rate(). Finally, Porcupine accepts input audio in consecutive chunks (aka frames) the length of each frame can be retrieved using pv_porcupine_frame_length().

extern const int16_t *get_next_audio_frame(void);

while (true) {
    const int16_t *pcm = get_next_audio_frame();
    int32_t keyword_index = -1;
    const pv_status_t status = pv_porcupine_process(handle, pcm, &keyword_index);
    if (status != PV_STATUS_SUCCESS) {
        // error handling logic
    }
    if (keyword_index != -1) {
        // Insert detection event callback
    }
}

Finally, when done be sure to release the acquired resources:

pv_porcupine_delete(handle);

Releases

v1.9.0 - December 2nd, 2020

• Added Alexa, Computer, Hey Google, Hey Siri, Jarvis, and Okay Google models under Apache 2.0.
• Added React Native SDK.
• Added Java SDK.
• Added .NET SDK.
• Added NodeJS SDK.
• Improved accuracy.
• Runtime optimization.

v1.8.0 - May 27th, 2020

• Improved accuracy.
• Runtime optimization.

v1.7.0 - Feb 13th, 2020

• Improved accuracy.
• Runtime optimization.
• Added support for Raspberry Pi 4.
• Added service-based Android demo application.
• Added C demo applications.
• Updated documentation.

v1.6.0 - April 25th, 2019

• Improved accuracy.
• Runtime optimization.
• Added support for BeagleBone.
• iOS build can run on simulator now.

v1.5.0 - November 13, 2018

• Improved accuracy.
• Runtime optimization.
• Added support for running within web browsers (WebAssembly).

v1.4.0 - July 20, 2018

• Improved accuracy.
• Runtime optimizations.
• Updated documentation.

v1.3.0 - June 19, 2018

• Improved accuracy.
• Runtime optimizations

v1.2.0 - April 21, 2018

• Runtime optimizations.

v1.1.0 - April 11, 2018

• Added multiple command detection capability.

v1.0.0 - March 13, 2018

• Initial release.

FAQ

You can find the FAQ here.