Eagle Speaker Recognition
Web API

API Reference for the Eagle Web SDK (npm).

Eagle

class Eagle {}

Class for using the recognizer component of the Eagle Speaker Recognition engine on the main application thread. The recognizer processes incoming audio in consecutive frames and emits a similarity score for each enrolled speaker.

Eagle.create()

static async function create(
  accessKey: string,
  model: EagleModel,
  options: EagleOptions = {}
): Promise<Eagle>

Creates an instance of the recognizer component of the Eagle Speaker Recognition engine.

Parameters

• accessKey string : AccessKey obtained from Picovoice Console.
• model EagleModel : Eagle model options.
• options EagleOptions: Optional configuration arguments:
  • device string : String representation of the device (e.g., CPU or GPU) to use. If set to best, the most suitable device is selected automatically. If set to gpu, the engine uses the first available GPU device. To select a specific GPU device, set this argument to gpu:${GPU_INDEX}, where ${GPU_INDEX} is the index of the target GPU. If set to cpu, the engine will run on the CPU with the default number of threads. To specify the number of threads, set this argument to cpu:${NUM_THREADS}, where ${NUM_THREADS} is the desired number of threads.
  • voiceThreshold number : Sensitivity threshold for detecting voice. The value should be a number within [0, 1]. A higher threshold increases detection confidence values at the cost of potentially missing frames of voice.

Returns

• Eagle : An instance of the Eagle.

Eagle.listAvailableDevices()

static async function listAvailableDevices(): Promise<string[]>

Lists all available devices that Eagle can use for inference. Each entry in the list can be the used as the device argument for the .create() method.

Returns

• string[] : List of all available devices that Eagle can use for inference.

Eagle.process()

async function process(
  pcm: Int16Array,
  speakerProfiles: EagleProfile[] | EagleProfile
): Promise<number[]>

Processes a frame of audio and returns a list of similarity scores for each speaker profile.

Parameters

• pcm Int16Array : A frame of audio samples. The minimum number of samples per frame can be attained by calling .minProcessSamples(). The incoming audio needs to have a sample rate equal to .sampleRate and be 16-bit linearly-encoded. Eagle operates on single-channel audio.
• speakerProfiles EagleProfile[] | EagleProfile : One or more Eagle speaker profiles. These can be constructed using EagleProfiler.

Returns

• number[] | null : A list of similarity scores for each speaker profile or null. A higher score indicates that the voice belongs to the corresponding speaker. The range is [0, 1] with 1 representing a perfect match. A result of null indicates that there was not enough voice in the audio to recognize any speakers.

Eagle.release()

async function release(): Promise<void>

Releases resources acquired by Eagle.

EagleProfiler.minProcessSamples

get minProcessSamples(): number

The minimum length of the input pcm required by .process().

Eagle.sampleRate

get sampleRate(): number

Audio sample rate accepted by Eagle.

Eagle.version

get version(): string

Version of Eagle.

EagleModel

type EagleModel = {
  base64?: string;
  publicPath?: string;
  customWritePath?: string;
  forceWrite?: boolean;
  version?: number;
}

Eagle model type.

• base64 string: The model file (.pv) in base64 string to initialize Koala.
• publicPath string: The model file (.pv) path relative to the public directory.
• customWritePath string : Custom path to save the model in storage. Set to a different name to use multiple models across Eagle instances.
• forceWrite boolean : Flag to overwrite the model in storage even if it exists.
• version number : Version of the model file. Increment to update the model file in storage.

EagleProfile

type EagleProfile = {
  bytes: Uint8Array;
}

Eagle speaker profile. Can be created by calling .export() after a successful speaker enrollment.

• bytes Uint8Array: Binary array containing the Eagle speaker profile.

EagleProfiler

class EagleProfiler {}

Class for using the profiler component of the Eagle Speaker Recognition engine on the main thread of your application. The profiler is responsible for enrolling a speaker given a set of utterances and exporting a speaker profile.

EagleProfiler.create()

static async function create(
  accessKey: string,
  model: EagleModel,
  options: EagleProfilerOptions = {}
): Promise<EagleProfiler>

Creates an instance of the profiler component of the Eagle Speaker Recognition engine.

Parameters

• accessKey string : AccessKey obtained from Picovoice Console.
• model EagleModel : Eagle model options.
• options EagleProfilerOptions: Optional configuration arguments:
  • device string : String representation of the device (e.g., CPU or GPU) to use. If set to best, the most suitable device is selected automatically. If set to gpu, the engine uses the first available GPU device. To select a specific GPU device, set this argument to gpu:${GPU_INDEX}, where ${GPU_INDEX} is the index of the target GPU. If set to cpu, the engine will run on the CPU with the default number of threads. To specify the number of threads, set this argument to cpu:${NUM_THREADS}, where ${NUM_THREADS} is the desired number of threads.
  • minEnrollmentChunks number : Minimum number of chunks to be processed before enroll returns 100%. The value should be a number greater than or equal to 1. A higher number results in more accurate profiles at the cost of needing more data to create the profile.
  • voiceThreshold number : Sensitivity threshold for detecting voice. The value should be a number within [0, 1]. A higher threshold increases detection confidence values at the cost of potentially missing frames of voice.

Returns

• EagleProfiler : An instance of the EagleProfiler.

EagleProfiler.enroll()

async function enroll(pcm: Int16Array): Promise<number>

Enrolls a speaker. This function should be called multiple times with different utterances of the same speaker until the enrollment percentage reaches 100.0 at which point a speaker voice profile can be exported using .export(). Any further enrollment can be used to improve the speaker voice profile. The number of required samples can be obtained by calling .frameLength. The audio data used for enrollment should satisfy the following requirements:

• only one speaker should be present in the audio
• the speaker should be speaking in a normal voice
• the audio should contain no speech from other speakers and no other sounds (e.g. music)
• it should be captured in a quiet environment with no background noise

Parameters

• pcm Int16Array : Audio data. The audio needs to have a sample rate equal to .sampleRate and be 16-bit linearly-encoded. EagleProfiler operates on single-channel audio.

Returns

• number : The percentage of enrollment completed.

EagleProfiler.flush()

async function flush(): Promise<number>

Marks the end of the audio stream, flushes internal state of the object, and returns the percentage of enrollment completed.

Returns

• number : The percentage of enrollment completed.

EagleProfiler.export()

async function export(): Promise<EagleProfile>

Exports the speaker profile of the current session. Will throw error if the profile is not ready.

Returns

• EagleProfile : The Eagle speaker profile.

EagleProfiler.reset()

async function reset(): Promise<void>

Resets the internal state of Eagle Profiler. It should be called before starting a new enrollment session.

EagleProfiler.release()

async function release(): Promise<void>

Releases resources acquired by Eagle Profiler.

EagleProfiler.frameLength

get frameLength(): number

Number of audio samples per frame expected by EagleProfiler (i.e. length of the array passed into .enroll())

EagleProfiler.sampleRate

get sampleRate(): number

Audio sample rate accepted by Eagle.

EagleProfiler.version

get version(): string

Version of Eagle.

EagleProfilerWorker

class EagleProfilerWorker {}

Class for using the profiler component of the Eagle Speaker Recognition engine on a worker thread. The profiler is responsible for enrolling a speaker given a set of utterances and exporting a speaker profile.

EagleProfilerWorker.create()

static async function create(
  accessKey: string,
  model: EagleModel,
  options: EagleProfilerOptions = {}
): Promise<EagleProfilerWorker>

Creates an instance of the profiler component of the Eagle Speaker Recognition engine.

Parameters

• accessKey string : AccessKey obtained from Picovoice Console.
• model EagleModel : Eagle model options.
• options EagleProfilerOptions: Optional configuration arguments:
  • device string : String representation of the device (e.g., CPU or GPU) to use. If set to best, the most suitable device is selected automatically. If set to gpu, the engine uses the first available GPU device. To select a specific GPU device, set this argument to gpu:${GPU_INDEX}, where ${GPU_INDEX} is the index of the target GPU. If set to cpu, the engine will run on the CPU with the default number of threads. To specify the number of threads, set this argument to cpu:${NUM_THREADS}, where ${NUM_THREADS} is the desired number of threads.
  • minEnrollmentChunks number : Minimum number of chunks to be processed before enroll returns 100%. The value should be a number greater than or equal to 1. A higher number results in more accurate profiles at the cost of needing more data to create the profile.
  • voiceThreshold number : Sensitivity threshold for detecting voice. The value should be a number within [0, 1]. A higher threshold increases detection confidence values at the cost of potentially missing frames of voice.

Returns

• EagleProfilerWorker : An instance of the EagleProfilerWorker.

EagleProfilerWorker.enroll()

async function enroll(pcm: Int16Array): Promise<number>

Enrolls a speaker. This function should be called multiple times with different utterances of the same speaker until the enrollment percentage reaches 100.0 at which point a speaker voice profile can be exported using .export(). Any further enrollment can be used to improve the speaker voice profile. The number of required samples can be obtained by calling .frameLength. The audio data used for enrollment should satisfy the following requirements:

• only one speaker should be present in the audio
• the speaker should be speaking in a normal voice
• the audio should contain no speech from other speakers and no other sounds (e.g. music)
• it should be captured in a quiet environment with no background noise

Parameters

• pcm Int16Array : Audio data. The audio needs to have a sample rate equal to .sampleRate and be 16-bit linearly-encoded. EagleProfilerWorker operates on single-channel audio.

Returns

• number : The percentage of enrollment completed.

EagleProfilerWorker.flush()

async function flush(): Promise<number>

Marks the end of the audio stream, flushes internal state of the object, and returns the percentage of enrollment completed.

Returns

• number : The percentage of enrollment completed.

EagleProfilerWorker.export()

async function export(): Promise<EagleProfile>

Exports the speaker profile of the current session. Will throw error if the profile is not ready.

Returns

• EagleProfile : The Eagle speaker profile.

EagleProfilerWorker.reset()

async function reset(): Promise<void>

Resets the internal state of Eagle Profiler. It should be called before starting a new enrollment session.

EagleProfilerWorker.release()

async function release(): Promise<void>

Releases resources acquired by Eagle Profiler.

EagleProfilerWorker.terminate()

async function terminate(): Promise<void>

Force terminates the instance of EagleProfilerWorker.

EagleProfilerWorker.frameLength

get frameLength(): number

Number of audio samples per frame expected by Eagle (i.e. length of the array passed into .enroll())

EagleProfilerWorker.sampleRate

get sampleRate(): number

Audio sample rate accepted by Eagle.

EagleProfilerWorker.version

get version(): string

Version of Eagle.

EagleWorker

class EagleWorker {}

Class for using the recognizer component of the Eagle Speaker Recognition engine on a worker thread. The recognizer processes incoming audio in consecutive frames and emits a similarity score for each enrolled speaker.

EagleWorker.create()

static async function create(
  accessKey: string,
  model: EagleModel,
  options: EagleOptions = {}
): Promise<EagleProfiler>

Creates an instance of the recognizer component of the Eagle Speaker Recognition engine.

Parameters

• accessKey string : AccessKey obtained from Picovoice Console.
• model EagleModel : Eagle model options.
• options EagleOptions: Optional configuration arguments:
  • device string : String representation of the device (e.g., CPU or GPU) to use. If set to best, the most suitable device is selected automatically. If set to gpu, the engine uses the first available GPU device. To select a specific GPU device, set this argument to gpu:${GPU_INDEX}, where ${GPU_INDEX} is the index of the target GPU. If set to cpu, the engine will run on the CPU with the default number of threads. To specify the number of threads, set this argument to cpu:${NUM_THREADS}, where ${NUM_THREADS} is the desired number of threads.
  • voiceThreshold number : Sensitivity threshold for detecting voice. The value should be a number within [0, 1]. A higher threshold increases detection confidence values at the cost of potentially missing frames of voice.

Returns

• EagleWorker : An instance of the EagleWorker.

EagleWorker.process()

async function process(
  pcm: Int16Array,
  speakerProfiles: EagleProfile[] | EagleProfile
): Promise<number[] | null>

Processes a frame of audio and returns a list of similarity scores for each speaker profile.

Parameters

• pcm Int16Array : A frame of audio samples. The minimum number of samples per frame can be attained by calling .minProcessSamples. The incoming audio needs to have a sample rate equal to .sampleRate and be 16-bit linearly-encoded. Eagle operates on single-channel audio.
• speakerProfiles EagleProfile[] | EagleProfile : One or more Eagle speaker profiles. These can be constructed using EagleProfiler.

Returns

• number[] | null : A list of similarity scores for each speaker profile or null. A higher score indicates that the voice belongs to the corresponding speaker. The range is [0, 1] with 1 representing a perfect match. A result of null indicates that there was not enough voice in the audio to recognize any speakers.

EagleWorker.release()

async function release(): Promise<void>

Releases resources acquired by EagleWorker

EagleWorker.terminate()

async function terminate(): Promise<void>

Force terminates the instance of EagleWorker.

EagleWorker.minProcessSamples

get minProcessSamples(): number

The minimum length of the input pcm required by .process().

EagleWorker.sampleRate

get sampleRate(): number

Audio sample rate accepted by Eagle.

EagleWorker.version

get version(): string

Version of Eagle.