Keyword detection - NN
Keyword detection - NN
free
not available
The following signal processing techniques are applied1. Recurrent Neural Network for training and decoding
2. LPC(Levinson Durbin) for time-domain feature extraction
3. Matrix norm regularization to avoid gradient explosion (common problem in RNN training)
4. Expectation and Maximization like VAD labeller (for onset and offset estimation)
Most variables are in Double precision (8bytes).Signal acquision is in 16kHz, 16bit PCM.
Overall, signal processing algorithms are thoroughly tested using Eclipse Java IDE. However, signal acquisition and app graphical design, layout design are still working
2. LPC(Levinson Durbin) for time-domain feature extraction
3. Matrix norm regularization to avoid gradient explosion (common problem in RNN training)
4. Expectation and Maximization like VAD labeller (for onset and offset estimation)
Most variables are in Double precision (8bytes).Signal acquision is in 16kHz, 16bit PCM.
Overall, signal processing algorithms are thoroughly tested using Eclipse Java IDE. However, signal acquisition and app graphical design, layout design are still working