VoiceNet Software system for automatic identification of speakers in telephone channels

The system is designed for automatic identification of speakers in telephone calls recordings.
VoiceNet automatically compares pairs of voice samples and gives out match probability.

APPLICATIONS

Police and other law enforcement agencies: Identification of suspected criminals by voice.

Anti-terrorist organizations: Identifying voices of terrorist recordered from phone calls.

Forensic labs: Storage and quick identification of speech samples.

HIGHLIGHTS

• 2 independent identification methods:
  • spectrum-formant (FFT)
  • pitch curve statistics comparison
• Effective for real signals, containing natural speech and telephone channel distortions
• Works with low quality speech signals
• Automatic identification does not require special user skills
• Language-independent (the speaker’s voice can be identified in any language he speaks)
• Standard DBMS allows external database import
• Any additional personal data (photos, fingerprints etc.) can be stored in the same database
• Available as SDK for software developers

SYSTEM DELIVERY

VoiceNet system includes one or more PCs connected via local network (LAN/WAN). The software consists of several independent modules located on client and server PCs:

• SpeechMarker – speech signal segmentation software
• FormBuilder – database creation, structure editing and access management software
• VoiceNetServer – database server and processing application
• VoiceNetOperator – database client software

SOURCE SIGNAL REQUIREMENTS

• Signal format - 16 bit PCM
• 8 000 or 11 025 Hz sampling rate
• Signal frequency band - 300...3400 Hz or better
• 10 dB or better SNR
• Speech duration for pair comparison - at least 16 seconds for each fragment

NEW FEATURES

• Additional speaker identification method – by pitch statistics
• New DB import/export tools for convenient data exchange
• User adjustable graphical interface
• Faster calculations and result output