Skip to content

Picovoice/picovoice-arduino-es

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Picovoice SDK for Arduino boards - Spanish language

Made in Vancouver, Canada by Picovoice

Picovoice is an end-to-end platform for building voice products on your terms. It enables creating voice experiences similar to Alexa and Google. But it entirely runs 100% on-device. Picovoice is

  • Private: Everything is processed offline. Intrinsically HIPAA and GDPR compliant.
  • Reliable: Runs without needing constant connectivity.
  • Zero Latency: Edge-first architecture eliminates unpredictable network delay.
  • Accurate: Resilient to noise and reverberation.

Compatibility

Dependency

  • LibPrintf

AccessKey

The Picovoice SDK requires a valid AccessKey at initialization. AccessKeys act as your credentials when using Picovoice SDKs. You can create your AccessKey for free. Make sure to keep your AccessKey secret.

To obtain your AccessKey:

  1. Login or Signup for a free account on the Picovoice Console.
  2. Once logged in, go to the AccessKey tab to create one or use an existing AccessKey.

Integration

  1. define all the necessary variables before setup():
#include <Picovoice_ES.h>

static const char* ACCESS_KEY = ... //AccessKey string obtained from [Picovoice Console](https://picovoice.ai/console/)

pv_picovoice_t *handle = NULL;

#define MEMORY_BUFFER_SIZE ...
static uint8_t memory_buffer[MEMORY_BUFFER_SIZE] __attribute__((aligned(16));

static const uint8_t *KEYWORD_ARRAY = ...
static const float PORCUPINE_SENSITIVITY = 0.75f;

static void wake_word_callback(void) {
    // logic to execute upon detection of wake word
}

static const uint8_t *CONTEXT_ARRAY = ...
static const float RHINO_SENSITIVITY = 0.5f;
static const float RHINO_ENDPOINT_DURATION_SEC = 1.0f;
static const bool RHINO_REQUIRE_ENDPOINT = true;

static void inference_callback(pv_inference_t *inference) {
    // `inference` exposes three immutable properties:
    // (1) `IsUnderstood`
    // (2) `Intent`
    // (3) `Slots`
    // ..
    pv_inference_delete(inference);
}

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 increased false alarm rate.

handle is an instance of Picovoice runtime engine that detects utterances of wake phrase defined in keyword_array. Upon detection of wake word it starts inferring user's intent from the follow-on voice command within the context defined in context_array. wake_word_callback is invoked upon the detection of wake phrase and inference_callback is invoked upon completion of follow-on voice command inference.

  1. put the following code block inside setup() in order to initialize the Picovoice engine:
const pv_status_t status = pv_picovoice_init(
            ACCESS_KEY,
            MEMORY_BUFFER_SIZE,
            memory_buffer,
            sizeof(KEYWORD_ARRAY),
            KEYWORD_ARRAY,
            PORCUPINE_SENSITIVITY,
            wake_word_callback,
            sizeof(CONTEXT_ARRAY),
            CONTEXT_ARRAY,
            RHINO_SENSITIVITY,
            RHINO_ENDPOINT_DURATION_SEC,
            RHINO_REQUIRE_ENDPOINT,
            inference_callback,
            &handle);

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

Picovoice accepts single channel, 16-bit PCM audio. The sample rate can be retrieved using pv_sample_rate(). Picovoice accepts input audio in consecutive chunks (aka frames); the length of each frame can be retrieved using pv_porcupine_frame_length(). Inside the loop() function in the sketch, pass the recorded audio to the Picovoice engine:

const int16_t *pcm = pv_audio_rec_get_new_buffer()
const pv_status_t status = pv_picovoice_process(handle, pcm);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

Create Custom Models

  1. Compile and upload the Picovoice_ES/GetUUID sketch from the File -> Examples menu. Copy the UUID of the board printed at the beginning of the session to the serial monitor.
  2. Go to Picovoice Console to create models for Porcupine wake word engine and Rhino Speech-to-Intent engine.
  3. Select Arm Cortex M as the platform when training the model.
  4. Select your board type (Arduino Nano 33 BLE Sense or Arduino Portenta H7) and provide the UUID of the chipset on the board.

The model is now being trained. You will be able to download it within a few hours.

Import the Custom Models

  1. Download your custom voice model(s) from Picovoice Console.
  2. Decompress the zip file. The model file is either .ppn for Porcupine wake word or .rhn for Rhino Speech-to-Intent. Both zip archives also contain a .h header file containing the C array version of the binary model.
  3. Copy the contents of the arrays inside the .h header files and update the KEYWORD_ARRAY (Porcupine) and CONTEXT_ARRAY (Rhino) values in params.h.

About

Picovoice SDK for Arduino boards - Spanish language

Resources

License

Stars

Watchers

Forks

Packages

No packages published