Model Index

Welcome to the model index of the Hiya Audio Intelligence APIs. Here you will find detailed descriptions and specifications of the models available in our API.

Check the guides for step by step flows to guide you through the most relevant operations.

Check the API reference for detailed descriptions of the endpoints, resources and operations.

Models are managed by Hiya and power the extraction and analysis of the characteristics of the speech data uploaded to the platform.

Voiceprint models

Extractors of the biometric characteristics of the voice.

Verification models

Analyzers and comparators of the characteristics of the speech data.

Voiceprint Models

Voiceprint models are extractors of the biometric characteristics of the voice. They can be used to compute voiceprints.

Model details: to review the requirements of a model navigate to the pages of its versions.

Default (v1)

The default voiceprint model is the only voiceprint model currently available, with more coming soon. The telephone channel and the English and Spanish languages make up a significant part of its training dataset.

Reference

Handle	default
Version	v1

Version selection: the platform automatically selects the latest versions of verification models.

Requirements

Audio

Maximum duration	2 minutes
Minimum voice duration	4.5 seconds / minimum audios
Minimum frequency	8000 Hz

Voiceprint

Maximum total duration	2 minutes
Minimum total voice duration	4.5 seconds
State	Computable

Verification Models

Verification models are analyzers and comparators of the biometric characteristics of the voice. They can be used to perform verifications.

Identity Verification

Identity verification models are analyzers and comparators of the biometric characteristics of the voice. They can be used to perform identity verifications.

Model details: to review the requirements of a model navigate to the pages of its versions.

Model selection: the platform automatically selects appropriate identity verification models for the chosen voiceprints. Every voiceprint model has a matching identity verification model.

Version selection: the platform automatically selects the latest versions of verification models.

Default (v1)

The default identity verification model is the only identity verification model currently available, with more coming soon. The telephone channel and the English and Spanish languages make up a significant part of its training dataset.

Reference

Handle	default
Version	v1.0

Requirements

Audio

Maximum duration	2 minutes
Minimum voice duration	1.5 seconds
Minimum frequency	8000 Hz
File state	Available

Voiceprint

Model	default
Model version	v1
State	Computed

Authenticity Verification

Authenticity verification models are analyzers of the authenticity and liveliness of the voice. They can be used to perform authenticity verifications.

Model details: to review the requirements of a model navigate to the pages of its versions.

Version selection: the platform automatically selects the latest version of the verification models.

Default (v10)

The default authenticity verification model is the recommended model for everything but the telephone channel. The digital channel and the English and Spanish languages make up a significant part of its training dataset.

v10 is the only version currently available and the previous ones have been deprecated, which does not impact in any way the verifications already performed with them.

Reference

Handle	default
Version	v10
Chunk duration	4 seconds

Requirements

Audio

Maximum duration	5 minutes
Minimum frequency	16000 Hz
File state	Available

Audio chunks

Minimum voice duration	750 milliseconds

Telephone (v6)

The telephone authenticity verification model is the recommended model for the telephone channel and is the only model that supports audios of 8 KHz. The English and Spanish languages make up a significant part of its training dataset.

v6 is the only version currently available.

Reference

Handle	telephone
Version	v6
Chunk duration	4 seconds

Requirements

Audio

Maximum duration	5 minutes
Minimum frequency	8000 Hz
File state	Available

Audio chunks

Minimum voice duration	750 milliseconds

Message Verification

Message verification models are analyzers and comparators of the messages found with the ones expected. They can be used to perform message verifications.

Model details: to review the requirements of a model navigate to the pages of its versions.

Version selection: the platform automatically selects the latest versions of verification models.

Default (v2)

The default message verification model is the only message verification model currently available. It supports a wide range of languages and expects a single phrase.

v2 is the only version currently available and the previous ones have been deprecated, which does not impact in any way the verifications already performed with them.

Reference

Handle	default
Version	v2

Requirements

Audio

Maximum duration	30 seconds
Minimum voice duration	2.5 seconds
Minimum frequency	8000 Hz
File state	Available

Message

Maximum length	200 characters
Supports digits	No

Language

Must be one of the following:
en, zh, de, es, ru, ko, fr, ja, pt, tr, pl, ca, nl, ar, sv, it, id, hi, fi, vi, he, uk, el, ms, cs, ro, da, hu, ta, no, th, ur, hr, bg, lt, la, mi, ml, cy, sk, te, fa, lv, bn, sr, az, sl, kn, et, mk, br, eu, is, hy, ne, mn, bs, kk, sq, sw, gl, mr, pa, si, km, sn, yo, so, af, oc, ka, be, tg, sd, gu, am, yi, lo, uz, fo, ht, ps, tk, nn, mt, sa, lb, my, bo, tl, mg, as, tt, haw, ln, ha, ba, jw, su or yue.