MovingSumWithSigmoid

Description

Performs MovingSum on the input alpha, and Sigmoid computation on the input energy. The product of the two results is the output.

Input

• alpha: input tensor of type float16 or float, with a shape of [bst].
• energy: input tensor of type float16 or float, with a shape of [bst].
• offset: input tensor of type int32, with a shape of [2 * batch_size].
• dec_data: input tensor of type float16 or float, with a shape of [bs, feature_dim].

Attribute

ksize: int.

Output

y: output tensor of type float16 or float, with a shape of [bs, bt].

Constraints

This operator is customized for specific user scenarios. You are advised not to use this operator in irrelevant scenarios.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16.