Porcupine Wake Word Engine - Microcontroller Quick Start

Platforms

• Arm Cortex-M4
• Arm Cortex-M7

Requirement

• C99-compatible compiler

Picovoice Account & AccessKey

1. Login or signup for a free account on the Picovoice Console.
2. Go to the AccessKey tab to create one or use an existing AccessKey. Be sure to keep your AccessKey secret.

Quick Start

Setup

1. Clone the repository:

git clone --recurse-submodules https://github.com/Picovoice/porcupine.git

Usage

1. Include the public header files.
2. Link the project to an appropriate library file.
3. Construct the porcupine object:

#define MEMORY_BUFFER_SIZE ${MEMORY_BUFFER_SIZE}
static const char* ACCESS_KEY = "${ACCESS_KEY}";

uint8_t memory_buffer[MEMORY_BUFFER_SIZE] __attribute__((aligned(16)));

const uint8_t keyword_array[] = ${KEYWORD_ARRAY};

const int32_t keyword_model_sizes = sizeof(${KEYWORD_ARRAY});
const void *keyword_models = keyword_array;
const float sensitivity = 0.5f;

pv_porcupine_t *porcupine = NULL;

const pv_status_t status = pv_porcupine_init(
        ACCESS_KEY,
        MEMORY_BUFFER_SIZE,
        memory_buffer,
        1,
        &keyword_model_sizes,
        &keyword_models,
        &sensitivity,
        &porcupine);

if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

4. Pass in frames of audio to the pv_porcupine_process function:

extern const int16_t *get_next_audio_frame(void);

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

5. Release resources explicitly when done with Picovoice:

pv_porcupine_delete(porcupine);

Create Custom Keywords

1. Obtain the UUID of the chipset.
2. Go to Picovoice Console to create models for Porcupine wake word engine.
3. Select Arm Cortex-M as the platform when training the model.
4. Select appropriate board type.
5. Train your models.
6. Download your custom keyword model.
7. Decompress the zip file. The .ppn file inside is the binary model for the Porcupine wake word. The zip archive also contains a .h header file containing the C array version of the binary model.
8. Copy the contents of the arrays inside the .h header files and update the keyword_array.

Non-English Languages

Use the corresponding library file (.a) to process non-English wake words. The library files for all supported languages are available on the Porcupine GitHub repository.

Demo

For the Porcupine mcu SDK, we offer demo projects for several evaluation boards to demonstrate how to use the Porcupine wake word engine on microcontrollers. The full list of supported boards are available here.

Setup

Clone the repository:

git clone --recurse-submodules https://github.com/Picovoice/porcupine.git

Usage

• IMXRT1050-EVKB (Arm Cortex-M7)
• PSoC6 (Arm Cortex-M4)
• STM32F407G-DISC1 (Arm Cortex-M4)
• STM32F411E-DISCO (Arm Cortex-M4)
• STM32F469I-DISCO (Arm Cortex-M4)
• STM32F769I-DISCO (Arm Cortex-M7)
• STM32H735G-DK (Arm Cortex-M4)
• STM32H747I-DISCO (Arm Cortex-M7)

Resources

API

• Microcontroller API doc

GitHub

• Porcupine mcu Demos on GitHub

Benchmark

• Wake Word Benchmark

Videos

• Offline voice recognition on MCUs - walkthrough