Information processing device, information processing method, information processing system, and program

A computer-implemented information processing method comprising:

• performing training processing of generating training data by convolving a measurement signal representing acoustic characteristics collected by a same sound collection unit as a sound collection unit used for collecting an observation signal with a reference signal having sound quality and reverberation characteristics different from the observation signal;
• generating teaching data with respect to the reference signal, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the measurement signal with the reference signal;
• and
• training a learning model for performing reverberation processing of the observation signal collected by the sound collection unit by using the training data and the teaching data as input data, wherein
• the direct sound component is estimated from the measurement signal collected in an anechoic room, and
• the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the reference signal.

A non-transitory computer readable memory having computer code stored therein that when executed by a processor causes the processor to perform a method, the method comprising:

• generating training data by convolving a measurement signal representing acoustic characteristics collected by a same sound collection unit as a sound collection unit used for collecting an observation signal with a reference signal having sound quality and reverberation characteristics different from the observation signal;
• generating teaching data with respect to the reference signal, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the measurement signal with the reference signal;
• and
• training a learning model for performing reverberation processing of the observation signal collected by the sound collection unit by using the training data and the teaching data as input data, wherein
• the direct sound component is estimated from the measurement signal collected in an anechoic room, and
• the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the reference signal.

An information processing device comprising:

• circuitry including a central processing unit (CPU) configured to
• input second acoustic data collected by a microphone and output third acoustic data obtained by removing a reverberation component from the second acoustic data by using a learning model obtained by performing machine learning by using first data generated by convolving first acoustic data collected by the microphone with dry data containing no reverberation and second data generated with respect to the dry data, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the first acoustic data with the dry data wherein the direct sound component is estimated from the first acoustic data collected in an anechoic room, and wherein the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the dry data,
• extract feature data representing a reverberation component of the second acoustic data by using the second acoustic data and the third acoustic data, and
• input sound source data, add reverberation having characteristics based on the feature data, and output the sound source data.

An information processing method comprising:

• causing a computer to execute processing of
• inputting second acoustic data collected by a microphone and outputting third acoustic data obtained by removing a reverberation component from the second acoustic data by using a learning model obtained by performing machine learning by using first data generated by convolving first acoustic data collected by the microphone with dry data containing no reverberation and second data generated with respect to the dry data, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the first acoustic data with the dry data, wherein the direct sound component is estimated from the first acoustic data collected in an anechoic room, and wherein the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the dry data,
• extracting feature data representing a reverberation component of the second acoustic data by using the second acoustic data and the third acoustic data, and
• inputting sound source data, adding reverberation having characteristics based on the feature data, and outputting the sound source data.

A non-transitory computer readable memory having computer code stored therein that when executed by a processor causes the processor to perform a method, the method comprising:

• inputting second acoustic data collected by a microphone and outputting third acoustic data obtained by removing a reverberation component from the second acoustic data by using a learning model obtained by performing machine learning by using first data generated by convolving first acoustic data collected by the microphone with dry data containing no reverberation and second data generated with respect to the dry data, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the first acoustic data with the dry data, wherein the direct sound component is estimated from the first acoustic data collected in an anechoic room, and wherein the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the dry data;
• extracting feature data representing a reverberation component of the second acoustic data by using the second acoustic data and the third acoustic data;
• and
• inputting sound source data, adding reverberation having characteristics based on the feature data, and outputting the sound source data.

An information processing system comprising:

• a microphone;
• a learning model that is obtained by performing machine learning on a computer by using first data generated by convolving first acoustic data collected by the microphone with dry data containing no reverberation and second data generated with respect to the dry data, from an average level and a delay value of a convolution signal generated by convolving a direct sound component of the first acoustic data with the dry data, wherein the direct sound component is estimated from the first acoustic data collected in an anechoic room, and wherein the delay value is a delay value of a convolution signal generated by extracting the estimated direct sound component and convolving the direct sound component with the dry data;
• a sound source that outputs sound source data;
• a memory that stores second acoustic data collected by the microphone and the sound source data;
• and
• circuitry including a central processing unit (CPU) coupled to the microphone and configured to
• input the second acoustic data from the microphone and output third acoustic data obtained by removing a reverberation component from the second acoustic data by using the learning model,
• extract feature data representing a reverberation component of the second acoustic data by using the second acoustic data and the third acoustic data, and
• input the sound source data, add reverberation having characteristics based on the feature data, and output the sound source data.