Speech Recognition

Robust even under difficult acoustic conditions

Intelligent speech input systems are increasingly being used as a user interface in human-technology interaction. The Branch Hearing, Speech and Audio Technology is working on speech detection systems that function robustly even under conditions with background noise and with a large distance between the microphone and speaker. The scientists use current scientific findings from psychoacoustic and psychophysical fundamental research to develop algorithms with a minimum false recognition rate. The phonologically-based speech recognisers can even be adapted to the reduced articulation of stroke patients.

Signal processing in embedded systems

When recogniser systems are implemented on microprocessors, all processing tasks are performed in the device and an internet connection is not needed. This also permits realisation of applications with high data protection

• Robust speech recognition in far-field and acoustically difficult environments
• Application-specific adaptation of vocabulary
• Microphoning and signal preprocessing
• Implementation of recogniser algorithms in embedded systems
• Evaluation studies