(Combination of Speaker, Speech, and Face Recognition with a single interface)
(Language- and Text-Independent, aka: Speaker Biometrics, Voice Biometrics, or SIV)
Recipient: Frost & Sullivan Award 2011
Embedded Speaker Recognition
Language- and Text-Independence: The speaker recognition system is completely text- and language-independent. This means that a user may enroll her/his voice into the system in one language and be identified or verified in a completely different language. This allows the engine to be able to handle authentication and identification processes across any number of languages.
The RecoMadeEasy® Speaker Recognition (SPKR) (SIV) System
is an award-winning engine developed entirely by Recognition
Technologies, Inc. which
currently runs on Linux, Mac, and Windows operating systems.
The engine is compatible with all microphone devices and most audio file formats.
system is fully integrated with our IVR system which is
telephony T1 and E1 cards as well as their analog cards.
It may also be run in a stand-alone environment independent of our
IVR system in a telephony or non-telephony setting.
This is a state-of-the-art language and
text-independent speaker recognition system (voice biometrics system) which has been developed
to work in different environments. Large-Scale and Small-Scale
versions of this speaker identification and speaker verification
(SIV) engine have been developed over many years of research to work
in the telephony as well as stand-alone environments. This speaker
biometric engine may be customized to fit your exact needs including
special modifications to fit the operating environment in which
your related applications run. Our staff has been actively
involved in defining speaker recognition (speaker biometric or voice biometrics)
standards in the VoiceXML and ANSI communities by providing
detailed consultation to the VoiceXML and M1 committees involved
in defining the speaker verification and identification standards.
The RecoMadeEasy® SIV system operates in 6 different
- Speaker Identification (Open-Set and Closed-Set)
The speaker enrolls his voice with the system. The system trains for
this and other speakers' voices. Once the speaker returns, the system
only has to listen to the speaker and will be able to identify the
speaker's voice among the trained voices it has in the database. The
identification process returns an ID for the speaker. There are two
different identification approaches. The simpler one is called
Closed-Set Identification in which case the ID of the closest voice in
the database is returned. In this case, if the speaker is not in the
database there is a possibility of a mis-tagged ID since the closest
voice is the database is picked. The more sophisticated (but harder)
approach is called Open-Set Identification where the speaker may
be tagged with an ID from the database or if the speaker has not been
enrolled in the database, he is rejected as not-enrolled.
Our SIV engine supports both Open-Set and Closed-Set approaches.
- Speaker Verification
In this modality, again, the speaker has to enroll his voice. Once the
enrollment process is done (recording of about 30 seconds of speech and
obtaining a positive ID of the speaker), the speaker is added to the
database. When the speaker returns, he makes a claim of his identity.
He will also speak for a few seconds and the speaker's voice is matched
against the database. His identity is either authenticated or he is
rejected as an impostor. It is important to note that there are two
possible sources of error; 1. False Acceptance and 2. False Rejection.
A false acceptance error would happen if the individual is mistakenly
authenticated. This is the number that we should try to minimize in
more security conscious applications. There is a trade-off between
the false acceptance and false rejection. If we reduce the false
acceptance rate, it means that we are making the security tighter. This
will naturally increase the number of false-rejections. False rejections
could become annoying if they are not limited.
- Speaker Classification and Event Detection
This modality of the engine may be used to classify speakers into
groups such as gender groups (male/female/child). Language detection
may also be viewed as classification. Age group and many other
categories may also be used to perform speaker classification. This may
also be used to classify or detect events such as beeps, speech, horn,
auto noise, background noise, etc.
- Speaker Detection
This would be the case where a speaker is already enrolled in the
database and we would be trying to find the speaker among recordings or
in a live conversation.
- Speaker Tracking
In this case a speaker's voice is tracked through the conversation and
the tracking makes sure the speaker stays on-line.
- Speaker Segmentation
This would be used to segment the speech between two or more speakers in
The Engine May be Used in the Following Ways
- Standalone engine which may be run through the use of
command lines and system calls.
- Standalone engine which may be used through a very simple
C++ SDK and API. This would be most useful for integrating
the engine into current products and IVR systems.
- As a module of our RecoMadeEasy® IVR system.
- As a web service using our servers.
- As a web service using your own servers.
Supported Audio Interface
The following interfaces are natively supported. However, the speaker
recognition engine may be used with any audio interface as long as
the audio is passed to the engine through a third party software such
as your own IVR system or recording program. The engine may be used
in many different scenarios such as a web service, C++ API, and
- All Microphone devices
- All Major Audio File Formats
- All Dialogic JCT Telephony cards (T1 and Analog)
Supported Operating Systems
The RecoMadeEasy® Embedded Speaker
Recognigtion engine is available for the following operating
systems. The C++ SDK, command-line interface, and web
services may be used in any of the following systems:
Embedded Operating Systems:
- Android (aarch64 -- armv8a)(Latest)
- Ubuntu 18.04 Mate Linux (aarch64)(Latest)
An evaluation account for the hosted version of
the RecoMadeEasy® Speaker
Recognition software may be made available to interested
Large-Vocabulary Speech Recognition
Available for English, Mandarin, Arabic, and German
Also Available in Bilingual Mandarin-English, Arabic-English, and German-English
(Customizable domain full transcription ~ 240,000+ word vocabulary)
(face detection and recognition)
Interactive Voice Response (IVR)
(Graph-based logic, easily configured)
Automatic Language Proficiency Rating (ALPR)
(Multi-lingual automated language proficiency rating)
Status: Advanced Development Stage
Status: Research Stage