ENVM-BiLSTM

LSTM-based Model for Noise Level Estimation.

The model was trainned on different spectral and chroma features extracted from the DNLE dataset.

Features:

20 MFCC
20 delta-delta
1 spectral centroid
12 chromagram
12 chroma energy normalized (CENS)

Model Architecture

Layer Type	Number of Layers
LSTM	3
BiLSTM	3
LSTM	3
Regression	1

Results

The DNLE dataset was split into 90% for training and 10% for testing. The model was trained with an "adam" optimizer, a learning rate of 0.0005, and a batch size of 675.
The model achieved an RMSE of 4.58 on average with a minimum of 0.5 and a maximum of 14.4 (scale of 30.6dBA to 81.3dBA), and a standard deviation of 2.72 on the test dataset.

Citing

If you find this model useful in an academic setting please consider citing: (to be updated)

Contact