v3.1.0

• Upgrade model from version 2 to version 3.
• To adapt model, I changed input shape from (64, 720) to (64, 512).
• By contrast to model_v2, add 2 layers.
• It achieves a more pleasure result.