| Type
of input accepted |
• Fixed phrase (text dependent)
• Any spoken input in limited domains
• Text prompted
• Free Speech
|
| Average
voiceprint size |
20 KBytess |
| Basic
technology |
Hidden Markov Models
+ Neural Networks |
| Anti
speaker modelling |
World Model |
| Configurable
parameters |
•Verification
grammar
• Confidence Scores at sentence and word level
• Verification threshold
• Number of enrollment/verification turns
• Expected utterance |
| Adaptation |
Environment/application adaptation
of the World Model |
| Threshold set-up |
Off-line testing, field trial, default set-up |
| Password/pass
phrase definition |
Application selected / user selected
(special free text
enrollment verification grammar) |
| Password
domain |
Any (depends on the enrollment grammar) |
| Supported languages |
Italian, Castilian Spanish, French, German,
Brazilian, Portuguese,
U.K. English, U.S. English, Greek, Chilean, Argentinean, Swedish,
Catalan, Mexican, Dutch. |
| OS supported |
MS Windows 2000, NT, XP, Linux Red Hat 7.x |
| Interfaces |
• Loquendo API
• MRCP (for Client Server Architecture)
• Intel Dialogic Audio Source support |
|
Enrollment / Verification
/ Identification
|
| Interaction
flow |
Customisable (on an application basis) |
| Minimum
required phrase length |
No hard requirements: 3-5 sec and at least 3
repetitions for
enrollment and 1 for verification/identification recommended |
| Acoustic
quality check |
Through ASR confidence, rejection and recognition
results |
| Multi
turn and mixed password processing |
Yes |
| Knowledge
Verification |
Yes |
| Identification
Capacity |
Up to 50 simultaneously |
| CPU requirements |
30 real time channels on a Pentium IV 3.2 GHz
for enrollment and
verification, 6 real time channels for the identification
of 40 users |
| Memory
requirements |
15 MB (fixed) + 5 MB per verification channel
or + 20 MB
per channel for the identification of 40 users |