Signal processing device, signal processing method, and computer program product

1. A signal processing device comprising:

one or more processors configured to:

perform first-type signal processing on an input signal using a neural network, and output a first-type output signal;

convert the first-type output signal into a second-type output signal for calculating a first-type loss related to accuracy of second-type signal processing performed by another signal processing device;

calculate the first-type loss based on the second-type output signal and a correct signal; and

optimize parameters of the neural network based on the first-type loss.

2. The signal processing device according to claim 1 , wherein the one or more processors:

calculate the first-type loss based on the second-type output signal and a correct signal that is converted into the first-type loss,

calculate a second-type loss based on the first-type output signal and a correct signal that is not converted into the first-type loss, and

optimize the parameters based on the first-type loss and the second-type loss.

3. The signal processing device according to claim 2 , wherein, depending on progress of learning, the one or more processors vary a weight of at least one of the first-type loss and the second-type loss.

4. The signal processing device according to claim 1 , wherein the correct signal represents a signal that has been converted into the first-type loss.

5. The signal processing device according to claim 1 , wherein the one or more processors convert the first-type output signal by a conversion method specified by the other signal processing device.

6. The signal processing device according to claim 1 , wherein the one or more processors:

calculate statistical information of the second-type output signal; and

based on first-type statistical information calculated from a plurality of second-type output signals corresponding to a plurality of input signals used in learning and second-type statistical information calculated from a second-type output signal corresponding to an input signal used in inference, calculate reliability of the first-type output signal corresponding to the input signal used in inference.

7. A signal processing method comprising:

performing, by a signal processing device, first-type signal processing on an input signal using a neural network, and outputting a first-type output signal;

converting, by the signal processing device, the first-type output signal into a second-type output signal for calculating a first-type loss related to accuracy of second-type signal processing performed by another signal processing device;

calculating, by the signal processing device, the first-type loss based on the second-type output signal and a correct signal; and

optimizing, by the signal processing device, parameters of the neural network based on the first-type loss.

8. A computer program product having a non-transitory computer readable medium including programmed instructions, wherein the instructions, when executed by a computer, cause the computer to perform:

performing first-type signal processing on an input signal using a neural network, and outputting a first-type output signal;

converting the first-type output signal into a second-type output signal for calculating a first-type loss related to accuracy of second-type signal processing performed by another signal processing device that is different from the computer;

calculating the first-type loss based on the second-type output signal and a correct signal; and

optimizing parameters of the neural network based on the first-type loss.

9. A signal processing device comprising:

one or more processors configured to:

perform first-type signal processing on input signals using a neural network, and output a first-type output signals;

convert the first-type output signals into a second-type output signals for calculating a first-type loss related to accuracy of second-type signal processing performed by another signal processing device; and

control signal processing to be performed, according to the second-type output signals corresponding to the input signals used in learning of parameters of the neural network performed by using the first-type loss.