Acoustic signal separation device and acoustic signal separating method
In an acoustic signal separation device ( 1 ), a determination unit ( 6 ) determines whether or not components from a plurality of sound sources are mixed in each of acoustic signals of respective components regenerated by a signal regeneration unit ( 5 ), and when it is determined that a plurality of components is mixed, a series of processes by a feature value extraction unit ( 2 ), a data estimation unit ( 3 ), a data classification unit ( 4 ), and a signal regeneration unit ( 5 ) is repeatedly executed until acoustic signals of the components of the respective sound sources are regenerated.
1. An acoustic signal separation device comprising:
processing circuitry
to extract a feature value from an input signal including an acoustic signal in which one or more components are mixed;
to estimate classification data on a basis of the feature value extracted by using a deep neural network trained to estimate the classification data for association between components of at least one acoustic signal output from an identical sound source;
to classify the classification data estimated into pieces of data corresponding to acoustic signals of respective components;
to regenerate the acoustic signals of the respective components, on a basis of both the classification data classified for each of the components of the acoustic signals and the feature value extracted; and
to determine whether or not components from a plurality of sound sources are mixed in each of the acoustic signals of the respective components regenerated, wherein
when it is determined that the components from the plurality of sound sources are mixed, a series of processes of the extracting, the estimating, the classifying, and the regenerating by the processing circuitry is repeatedly executed until acoustic signals of the components of the respective sound sources are regenerated.
2. The acoustic signal separation device according to claim 1 , wherein
the processing circuitry determines whether or not the components from the plurality of sound sources are mixed on a basis of a speech recognition result of each of the acoustic signals of the respective components.
3. The acoustic signal separation device according to claim 1 , wherein
the processing circuitry determines whether or not the components from the plurality of sound sources are mixed on a basis of a recognition likelihood for speech recognition on each of the acoustic signals of the respective components.
4. An acoustic signal separating method comprising:
extracting a feature value from an input signal including an acoustic signal in which one or more components are mixed;
estimating classification data on a basis of the feature value extracted, by using a deep neural network trained to estimate the classification data for association between components of at least one acoustic signal output from an identical sound source;
classifying the classification data estimated into pieces of data corresponding to acoustic signals of respective components;
regenerating the acoustic signals of the respective components, on a basis of both the classification data classified for each of the components of the acoustic signals and the feature value extracted; and
determining whether or not components from a plurality of sound sources are mixed in each of the acoustic signals of the respective components regenerated, wherein
when it is determined that the components from the plurality of sound sources are mixed, a series of processes of the extracting, the estimating, the classifying, and the regenerating is repeatedly executed until acoustic signals of the components of the respective sound sources are regenerated.