Speaker recognition technology using voice is applied to biometric fields through utilization of not only language information included in voice, but also the information of speakers’ characteristics (voiceprint). In addition, speaker verification is applied as an essential technology to individual customization service in intelligent robot and smart electronic appliance fields.
Key algorithm
The algorithm to extract voice characteristic parameters by speaker using adaptive component weighting maximizes speaker distinction ability. In addition, a combined model VQ/ HMM speaker distinction algorithm to maximize biometric information and minimize voice information is developed. This is applied with speaker adaptation technology and a signal bias removal algorithm, and thus is developed into a speaker recognition system that is powerful in noise removal only with a small amount of learning data.
- Inter-Speaker Variation Minimization
- Algorithm to generate a variable-length code book for improved speaker distinction ability
- Combined structure speaker modeling technology using world model and cohort model
- Channel mismatch compensation algorithm to consider user’s mobility
- Technology for normalizing factors to change speaker’s characteristics, such as speaker’s physical condition and
surrounding environment - Speaker Adaptation
- Modeling technology to predict long-term changes in voice characteristics according to changes over time
- Text-dependent/ text-independent speaker recognition technology
- Context-dependent speaker recognition system where speakers pronounce a designated sentence
- User-selectable password speaker recognition system where speakers pronounce a password they have selected
- Unspecific context speaker recognition system where speakers' characteristics are extracted and recognized from
the pronunciation of a long sentence
