Research programs
Speech, emotion & human context
Learning from speech while keeping the speaker, the recording conditions and the limits of interpretation in view.
Research direction
How can computational models describe emotional variation in speech? This line of work studies raw-waveform neural architectures, temporal context and emotion annotation, including research with crisis-helpline recordings.
The publications connect convolutional networks, recurrent models and multiresolution signal analysis. Benchmark results describe performance within a study; they do not establish a clinical diagnostic system or a replacement for a counsellor’s judgment.
- What can a model learn directly from a variable-length waveform?
- How do evaluation conditions affect emotion recognition?