cry_analyzer

Model

https://drive.google.com/file/d/1szE_3IFeeR5-WK1p4GEbuyg1xlWRdH4O/view?usp=sharing

Results

Validation Accuracy: 38.0%

Model predicts 4 different classes:

• belly_pain : 11 predictions ( 13.9%)
• burping : 22 predictions ( 27.8%)
• discomfort : 10 predictions ( 12.7%)
• hungry : 36 predictions ( 45.6%)

Per-class Accuracy:

• belly_pain : 30.8% (4/13)
• burping : 76.9% (10/13)
• discomfort : 18.8% (3/16)
• hungry : 59.1% (13/22)
• tired : 0.0% (0/15)

Dataset

I borrowed from:

• Infant cry audio corpus

Since the dataset is too imbalanced, I ueed some augmentation techniques to balance the dataset.

• --target for the number of target data after balancing

python balance_data.py cry_data --target 60 --strategy hybrid

After balanced:

RasberryPi

Here