This document discusses the development of an emotional Kannada speech database and its analysis and evaluation for building an effective emotion recognition system. Key points:
- A database of 60 sentences in Kannada was created, expressing happiness, sadness, anger, fear and neutral emotions as uttered by 2 male actors.
- Acoustic features like pitch, intensity, percentage of unvoiced frames, sound pressure, vocal tract variations were analyzed for the different emotions using PRAAT software.
- Statistical analysis found pitch was highest in fear and lowest in sadness, while intensity was highest in anger and lowest in fear. Unvoiced frames were highest in fear and lowest in happy.
- Linear predictive coding (LPC) was