GritTec's Speaker-ID: Automatic Text Independent Speaker Identification
GritTec's Speaker-ID: Automatic Text Independent Speaker
Identification (Version 4.00) is intended for automatic voice identification or voice verification of a speech
signal of unknown speaker by paired comparing with speech signal of target speaker.
Designed algorithm of speaker identifications is based on duel
comparison spectra features of unknown voice with the spectra
features of target voice. Spectra features are calculated with
provision of dynamic determinations of channel distortion level and
external hindrances and noises.
It allows to compensate channel distortion and influences of external hindrances with comparing spectra features, put into the
original speech signal. Sensitivity to identifications is defined by the level of installing the thresholds of probability of errors 1-th
(False Rejection Rate (FRR)) and 2-th (False Acceptance Rate (FAR)) sort. Possibility of regulation of thresholds of FRR and FAR allows to adjust a process of identification flexibly in accordance with system safety requirements.
At the moment the GritTec's Speaker-ID engine are realized in software solution of voice identification
with GUI interface - GritTec Speaker-ID: The mobile client.
Fig. GritTec Speaker-ID: The mobile client.
- For automatic voice identification or verification of unknown voice by phonogram of telephone negotiations;
- In systems with high safety level, for instance, when
access to digital information is limited by circle of given
- Applications where it's necessary to identify a person using
peculiarities of his voice.
- Operation with low SNR;
- Fast adaptation to changing of channel distortion and external
- Minimum duration of a speech signal with a voice example used for correct reception of voice parameters for the target speaker - not less 15 seconds;
- Minimum duration of a speech signal with a voice example used for voice identification or voice verification - not less 7 seconds;
- Speaker identification reliability not less than 90% if both of speech signals were recorded in the same channel;
- Speaker identification reliability not less then 85% if both of speech signals were recorded in different channels (cross channels);
- Supporting voice identification or voice verification in multi-threading mode;
- Automatic voice identification or voice verification doesn't require special skills;
- Supporting the software license key or hardware license with USB key (see Abstract of GritTec's SDK delivery);
- Supporting the license key depends from the quantity of target samples are used in stream of voice identification;
- Supporting the license key depends from the quantity of voice identification streams or depends from the quantity of voice training streams;
- Easy integration with target applications.
- Signal format: 16-bits linear, 8 kHz sampling rate;
- SNR, at least 10 db;
- Frequency range: 300-3400 Hz or better.
- Demo programs under MS Windows x86/x64, Linux x86/x64 platforms;
- The software solution with GUI interface under MS Windows x86/x64 platforms
GritTec Speaker-ID: The mobile client;
- SDK packets for developers under MS Windows x86/x64, Linux x86/x64 platforms;
- Object ANSI C/C++ code;
- Java wrapper under Java SE (x86/64) Runtime Environment;
For an estimation accuracy of GritTec's Speaker-ID engine in mode of voice verification it was used voices of 25
target speakers (12 - males, 13 - females) for English language. Each target speaker was
trained separately for CELL and VOIP channel. It was used 50 files for training, 25 files - for CELL channel and
25 files - for VOIP channel. Each trained file contained 12 phrases
of a random digit numbers (from 0 to 5) with the common duration ~ (40-50) seconds.
The total of files used for verification in CELL and VOIP channels
was 31950, where 30195 files with voices of imposter speakers, and 1755 files with voices of target speakers. Each
verified file contained 1 phrase of a random digit numbers (from 0 to 6) with the common duration ~ (4-6) seconds.
Screenshots of DET curves and EER (Equal Error Rate) errors of verification results for CELL and VOIP channels are shown
EER: 5,96 % (training on CELL,
verification on CELL);
EER: 7.01 % (training on CELL, verification on
CELL and VOIP);
EER: 3.91 % (training on VOIP, verification on
EER: 8.11 % (training on VOIP, verification on
CELL and VOIP);
All results have been received for GritTec's Speaker-ID (Version 2,90)
with comparing Version 2,80.
Online Shopping and Free Downloads
- All products include a free evaluation period for your first installation only;
- To continue using these programs after your evaluation period has expired, you can buy a end-user license;
- Click 'Buy Now' under current product.
For more information, please contact us via Online Request Form.