Eagle Speaker Recognition Engine
C Quick Start

Platforms

• Linux (x86_64)
• macOS (x86_64, arm64)
• Windows (x86_64, arm64)
• Raspberry Pi (3, 4, 5)

Requirements

• C99-compatible compiler
• CMake (3.13+)
• For Windows Only: MinGW is required to build the demo

Picovoice Account & AccessKey

Signup or Login to Picovoice Console to get your AccessKey. Make sure to keep your AccessKey secret.

Overview

Eagle Speaker Recognition consists of two distinct steps: Enrollment and Recognition. In the enrollment step, Eagle analyzes a series of utterances from a particular speaker to learn their unique voiceprint. This step results in a Profile object, which can be stored and utilized during inference. During the Recognition step, Eagle compares the incoming frames of audio to the voiceprints of all enrolled speakers in real-time to determine the similarity between them.

Quick Start

Setup

1. Clone the repository:

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

Usage

1. Include the public header files (picovoice.h and pv_eagle.h).
2. Link the project to an appropriate precompiled library for the target platform and load it.

Speaker Enrollment

3. Construct an instance of the profiler:

const char *access_key = "${ACCESS_KEY}";
const char *model_path = "${MODEL_PATH}";
const char *device = "best";
const int32_t min_enrollment_chunks = 1;
const float voice_threshold = 0.3f;

pv_eagle_profiler_t *eagle_profiler = NULL;
pv_status_t status = pv_eagle_profiler_init(
            access_key,
            model_path,
            device,
            min_enrollment_chunks,
            voice_threshold,
            &eagle_profiler);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

4. Pass in audio to the pv_eagle_profiler_enroll function to enroll a speaker:

extern const int16_t *get_next_enroll_audio_frame(int32_t frame_length);
extern const bool has_next_enroll_audio_frame(int32_t frame_length);
const int32_t frame_length = pv_eagle_profiler_frame_length();

float enroll_percentage = 0.0f;

while (enroll_percentage < 100.0f && has_next_enroll_audio_frame(frame_length)) {
  status = pv_eagle_profiler_enroll(
          eagle_profiler,
          get_next_enroll_audio_frame(frame_length),
          &enroll_percentage);
  if (status != PV_STATUS_SUCCESS) {
      // error handling logic
  }
}

status = pv_eagle_profiler_flush(
        eagle_profiler,
        &enroll_percentage);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

int32_t profile_size_bytes = 0;
status = pv_eagle_profiler_export_size(eagle_profiler, &profile_size_bytes);
void *speaker_profile = malloc(profile_size_bytes);
status = pv_eagle_profiler_export(
        eagle_profiler,
        speaker_profile);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

5. Release resources explicitly when done with the profiler:

pv_eagle_profiler_delete(eagle_profiler);

Speaker Recognition

3. Construct an instance of the engine:

pv_eagle_t *eagle = NULL;
pv_status_t status = pv_eagle_init(
        access_key,
        model_path,
        device,
        voice_threshold,
        &eagle);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

4. Pass audio frames to the pv_eagle_process function to perform speaker recognition:

extern const int16_t *get_next_audio_frame(int32_t min_process_samples);

const int32_t min_process_samples = 0;
pv_status_t status = pv_eagle_process_min_audio_length_samples(
        eagle,
        &min_process_samples);
if (status != PV_STATUS_SUCCESS) {
    // error handling logic
}

float *scores = NULL;
while (true) {
    status = pv_eagle_process(
            eagle,
            get_next_audio_frame(min_process_samples),
            min_process_samples,
            &speaker_profile,
            1,
            &scores);
    if (status != PV_STATUS_SUCCESS) {
        // error handling logic
    }
}

5. Release resources explicitly when done with the engine:

pv_eagle_delete(handle);

Demo

For the Eagle Speaker Recognition SDK, we offer demo applications that demonstrate how to use the speaker recognition engine on real-time audio streams (i.e. microphone input) and audio files.

Setup

1. Clone the Eagle Speaker Recognition repository from GitHub using HTTPS:

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

2. Build the microphone demo:

cd eagle
cmake -S demo/c/ -B demo/c/build
cmake --build demo/c/build --target eagle_demo_mic

Usage

To see the usage options for the demo:

./demo/c/build/eagle_demo_mic

Ensure you have a working microphone connected to your system and run the command corresponding to your platform to either enroll a speaker or perform speaker recognition:

./demo/c/build/eagle_demo_mic \
-l lib/${PLATFORM}/${ARCH}/libpv_eagle.${LIB_EXTENSION} \
-m lib/common/eagle_params.pv \
-a ${ACCESS_KEY}
-e ${OUTPUT_PROFILE_PATH}

or

./demo/c/build/eagle_demo_mic \
-l lib/${PLATFORM}/${ARCH}/libpv_eagle.${LIB_EXTENSION} \
-m lib/common/eagle_params.pv \
-a ${ACCESS_KEY}
-i ${INPUT_PROFILE_PATH}

For more information on our Eagle Speaker Recognition demos for C, head over to our GitHub repository.

Resources