VeriSpeak SDK - Standard Edition

Voice / Speaker identification for PC and Web solutions

VeriSpeak voice identification technology is intended for biometric systems developers and integrators. The text-dependent speaker recognition algorithm assures system security by checking both voice and phrase authenticity. Voiceprint templates can be matched in 1-to-1 (verification) and 1-to-many (identification) modes.

VeriSpeak is available as a software development kit that allows development of PC- and Web-based solutions on Microsoft Windows, Linux and Mac OS X platforms.

Advantages of VeriSpeak

• Text-dependent algorithm prevents unathorized access with a covertly recorded user voice.
• Two-factor authentication by checking voice biometrics and pass phrase authenticity.
• Regular microphones are suitable for recording user voices.
• Available as a multiplatform SDK that supports multiple programming languages.
• Reasonable prices, flexible licensing and free customer support.

VeriSpeak Algorithm Features and Capabilities

All performance tests were made on Intel Core2 processor with 4 cores running at 2.66 GHz.

VeriSpeak is a PC-based speaker recognition algorithm designed for biometric system integrators. The VeriSpeak algorithm implements voice enrollment and voiceprint matching using proprietary sound processing technologies:

• Text-dependent algorithm. The text-dependent speaker recognition is based on saying the same phrase for enrollment and verification. The VeriSpeak algorithm determines if a voice sample matches the template that was extracted from a specific phrase. During enrollment, one or more phrases are requested from the person being enrolled. Later that person may be asked to pronounce a specific phrase for verification. This method assures protection against the use of a covertly recorded random phrase from that person.
• Two-factor authentication with a passphrase. The VeriSpeak voiceprint matching algorithm can be configured to work in a scenario, where each user records a unique phrase (such as passphrase or an answer to a "secret question" that is known only by the person being enrolled). Later a person is recognized by his or her own specific phrase with a high degree of accuracy. The overall system security increases as both voice authenticity and passphrase are checked.
• Liveness detection. A system may request each user to enroll a set of unique phrases. Later the user will be requested to say a specifc phrase from the enrolled set. This way the system can ensure that a live person is being verified (as opposed to an impostor who uses a voice recording).
• Identification capability. VeriSpeak functions can be used in 1-to-1 matching (verification) and 1-to-many (identification) modes.
• Multiple samples of the same phrase. A template may store several voice records with the same phrase to improve recognition reliability. Certain natural voice variations (i.e. hoarse voice) or environment changes (i.e. office and outdoors) can be stored in the same template.
• Fused matching. A system may ask users to pronounce several specific phrases during speaker verification or identification and match each audio sample against records in the database. The VeriSpeak algorithm can fuse the matching results for each phrase together to improve matching reliability.

VeriSpeak Technology and SDK

• VeriSpeak Standard and Extended SDK. VeriSpeak Standard SDK is intended for PC-based biometric application development and the Extended SDK is suitable for developing Web-based biometric systems. Both SDKs include Voice Matcher and Extractor components; the Extended SDK also includes Voice Client component.   Read more
• Recommendations and constraints for speaker recognition. VeriSpeak has certain requirements for microphone settings and position, as well as user behavior and environment. A passphrase should be kept in secret and not pronounced in an environment where other people may hear it.   Read more
• System requirements. VeriSpeak-based software can be run on computers with x86 compatible processors (at least 2 GHz processor recommended). Windows, Linux and Mac OS X platforms are supported. A regular microphone is suitable for voice capture.   Read more
• Technical specifications. VeriSpeak extracts a template in 0.07 seconds and can match up to 88 voiceprints per second, when 5-second long voice samples are used. A single voiceprint template requires about 6.5 kilobytes.   Read more
• Reliability and performance tests. The VeriSpeak template extraction and matching algorithm has been tested with the voice samples XM2VTS database and Neurotechnology internal databases. Read more
• Download. A VeriSpeak 30-day SDK Trial is available for downloading.

Basic Recommendations for Speaker Recognition

The speaker recognition accuracy of VeriSpeak and MegaMatcher depends on the audio quality during enrollment and identification. Certain constraints should be noted before or during algorithm integration into a speaker recognition system, whereas other can be overcome by enrollment with the same phrase in different environments.

At least 2-seconds long voice samples are recommended to assure recognition quality.

General Security

A passphrase should be kept in secret and not pronounced in an environment where other people may hear it if the speaker recognition system is used in a scenario with unique phrases for each user.

Microphones

There are no particular constraints on models or manufacturers when using regular PC microphones, headsets or built-in laptop microphones. However these factors should be noted:

• The same microphone model is recommended (if possible) for use during both enrollment and recognition, as different models can produce different sound quality. Also some models may introduce specific noise or distortion into the audio, or may include certain hardware sound processing, which will not be present when using a different model.
• The same microphone position and distance is recommended during enrollment and recognition. Headsets provide optimal distance between user and microphone; this distance is recommended when non-headset microphones are used.
• Web cam built-in microphones should be used with care, as they are usually positioned at a rather long distance from the user and may provide lower sound quality. The sound quality may be affected if users change their position relative to the web cam.

Sound Settings

Settings for clear sound must be ensured, as some audio software, hardware or drivers may have certain means of sound modification enabled by default. For example, the Microsoft Windows OS usually has sound boost enabled by default.

At least 11,025 Hz sampling rate with at least 16-bit depth should be set during voice recording.

Environment Constraints

The VeriSpeak and MegaMatcher speaker recognition algorithm is sensitive to background noise or loud voices in the background that may interfere with the user's voice and affect the recognition results. These solutions may be considered to reduce or eliminate these problems:

• A silent environment for enrollment and recognition.
• Several samples of the same phrase recorded in different environments can be stored in a biometric template. Later the user will be matched against these samples with much higher recognition quality.
• Close-range microphones (like those in headsets) that are not affected by distant sources of sound.
• Third-party or custom solutions for background noise reduction, like using two separate microphones for recording user voice and background sound, and later subtracting the background noise from the recording.

User Behavior and Voice Changes

These natural voice changes do not occur often but may affect speaker recognition accuracy:

• A temporarily hoarse voice caused by a cold or other sickness
• Different emotional states that affect voice (i.e. cheerful voice versus tired voice)
• Different pronuncation speeds during enrollment and identification

The aforementioned voice and user behavior changes can be managed in two ways:

• Separate enrollments for the altered voice with storing the records to the same person's template;
• Controlled neutral voice during enrollment and identification.

VeriSpeak SDK System Requirements

• PC or Mac with x86 (32-bit) or x86-64 (64-bit) compatible processors. 2GHz or better processor is recommended.
• At least 128 MB of free RAM should be available for the application. Additional RAM is required for applications that perform 1-to-many identification, as all biometric templates need to be stored in RAM for matching. For example, 1,000 templates (each containing 1 voice record) require about 6 MB of additional RAM.
• Free space on hard disk drive (HDD):
  • at least 1 GB required for the development.
  • 100 MB required for VeriSpeak components deployment.
  • Additional space would be required in these cases:
    • VeriSpeak does not require the original voice sample to be stored for the matching; only the templates need to be stored. However, storing voice samples on hard drive for the potential future usage is recommended.
    • Usually a database engine runs on a separate computer (back-end server). However, DB engine can be installed on the same computer for standalone applications. In this case HDD space for templates storage must be available. For example, 10,000 templates (each containing 1 voice record extracted from 5-seconds long sample) stored using a relational database would require about 64 MB of free HDD space. Also, the database engine itself requires HDD space for running. Please refer to HDD space requirements from the database engine providers.
• Microphone. Any microphone that is supported by the operating system can be used.
• Database engine or connection with it. VeriSpeak templates can be saved into any DB (including files) supporting binary data saving. VeriSpeak Extended SDK contains the following support modules for Matching Server:
  • Microsoft SQL Server (only for Microsoft Windows platform);
  • PostgreSQL (for Microsoft Windows and Linux platform);
  • MySQL (for Microsoft Windows and Linux platforms);
  • Oracle (for Microsoft Windows and Linux platforms);
  • SQLite (for all platforms).
• Network/LAN connection (TCP/IP) for client/server applications. Also, network connection is required for using Matching server component (included in VeriSpeak Extended SDK). Communication with Matching server is not encrypted therefore, if communication must be secured, a dedicated network (not accessible outside the system) or a secured network (such as VPN; VPN must be configured using operating system or third party tools) is recommended.
• Microsoft Windows specific requirements:
  • Microsoft Windows XP/Vista/7, 32-bit or 64-bit.
  • Microsoft .NET framework 2.0 or newer (for .NET components usage).
  • One of following development environments for application development:
    • Microsoft Visual Studio 2005 SP1 or newer (for application development under C/C++, C#, Visual Basic .Net)
    • Sun Java 1.5 SDK or later
    • Delphi 7
• Linux specific requirements:
  • Linux 2.6 or newer kernel, 32-bit or 64-bit.
  • glibc 2.7 or newer
  • GStreamer 0.10.23 (with gst-plugin-base and gst-plugin-good) or newer (for voice capture using microphone)
  • libasound 1.0.x or newer (for voice capture)
  • udev-143 or newer with libudev (for microphone usage)
  • GTK+ 2.10.x or newer libs and dev packages (to run SDK samples and applications based on them)
  • GCC-4.0.x or newer (for application development)
  • GNU Make 3.81 or newer (for application development)
  • Sun Java 1.5 SDK or later (for application development with Java)
  • pkg-config-0.21 or newer (optional; only for Matching Server database support modules compilation)
• Mac OS X specific requirements:
  • Mac OS X (version 10.4 or newer)
  • XCode 2.4 or newer (for application development)

Technical Specifications

At least 11,025 Hz sampling rate with at least 16-bit depth should be configured during voice recording.

At least 2-second long voice samples are recommended to assure recognition quality. Longer voice samples will improve the recognition quality.

See also the whole list of recommendations and constraints for speaker recognition.

All voice templates should be loaded into RAM before identification, thus the maximum voice template database size is limited by the amount of available RAM.

The VeriSpeak voice template matching algorithm can use more than one processor core on multi-core processors, enabling an increase in template matching speed. The template matching speeds in the table below are given as a range, where the smaller number means matching speed using 1 processor core, while the larger number means matching speed using all 4 processor cores.

The specifications below have linear dependence from the voice sample length. For example, when using voice samples that are 2 times shorter, template extraction will take 2 times less time, the extracted template will be 2 times shorter, and the templates will be matched 2 times faster.

VeriSpeak algorithm technical specifications (for 5 second long voice samples)
Voice template extraction time (milliseconds)	60 - 90
Matching speed (voiceprints per second)	22 - 88
Template size in database (1) (kilobytes)	6.0 - 6.8

(1) When 1 voiceprint record is stored in a template. Template size increases proportionally when multiple voiceprint records are stored in the same template.

Back to top of this page

Reliability and Performance Tests

All tests were performed on Intel Core2 processor with 4 cores running at 2.66 GHz.

The VeriSpeak algorithm has been tested with voice samples taken from the XM2VTS Database, as well as with voice samples from Neurotechnology internal database. These voice template extraction and matching experiments were performed: Experiment 1 – used voice samples from the XM2VTS database. All samples include the same phrase pronounced by all subjects.

Experiment 1 Click to zoom

Experiment 2 – used voice samples from Neurotechnology internal voice database 1. All samples include the same phrase pronounced by all subjects. Experiment 3 – used voice samples from Neurotechnology internal voice database 2. Each subject pronounced a unique phrase during his/her recording. Template matching was performed using all 4 cores of the processor. Receiver operation characteristics (ROC) curves are usually used to demonstrate the recognition quality of an algorithm. ROC curves show the dependence of false rejection rate (FRR) on the false acceptance rate (FAR). Charts with ROC curves for both databases are available on the right.

Experiments 2 and 3 Click to zoom

VeriSpeak algorithm tests with XM2VTS and Neurotechnology internal databases
	Exp. 1	Exp. 2	Exp. 3
Total voice samples in the database	2360	309	305
Subjects in the database	295	42	42
Recording sessions per subject	8	1 - 10	1 - 10
Average voice sample length (seconds)	7.112	4.975	6.214
Average template extraction speed (milliseconds)	132	56	69
Average voiceprint template size (kilobytes)	8.7	6.7	7.8
Template matching speed (voiceprints per second)	64	136	88

Back to top of this page