Method and apparatus for classifying signals, method and apparatus for generating descriptors and method and apparatus for retrieving signals

A method for classifying signals comprising:

• dividing an input signal into blocks having a predetermined time length;
• extracting one or more than one characteristic quantities of a signal attribute from the signal of each block;
• and
• classifying the signal of each block into a category according to the characteristic quantities thereof, wherein said signal of each block is classified into any of the categories formed on the basis of types structures that signals may have and do not depend on the types of signal sources, and wherein one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block are used as said characteristic quantities.

An apparatus for classifying signals comprising:

• a blocking means for dividing an input signal into blocks having a predetermined time length.
• a feature extracting means for extracting one or more than one characteristic quantities of a signal attribute from the signal of each block;
• and
• a categorical classifying means for classifying the signal of each block into a category according to the characteristic quantities thereof, wherein said categorical classifying means classifies said signal of each block into any of the categories formed on the basis of types of structures that signals may have and do not depend on the types of signal sources, and wherein said feature extracting means uses one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block as said characteristic quantities.

A method for generating descriptors comprising:

• dividing an input signal into blocks having a predetermined time length;
• extracting one ore more than one characteristic quantities of a signal attribute from the signal of each block;
• classifying the signal of each block into a category according to the characteristic quantities thereof, wherein said signal of each block is classified into any of the categories formed on the basis of types of structures that signals may have and do not depend on the types of signal sources, and
• generating a descriptor for the signal according to the category of classification thereof,
• wherein one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block are used as said characteristic quantities.

An apparatus for generating descriptors comprising:

• a blocking means for dividing an input signal into blocks having a predetermined time length:
• a feature extracting means for extracting one or more than one characteristic quantities of a signal attribute from the signal of each block;
• a categorical classifying means for classifying the signal of each block into a category according to the characteristic quantities thereof wherein said categorical classifying means classifies said signal of each block into any of the categories formed on the basis of types of structures that signals may have and do not depend on the types of signal sources;
• and
• a descriptor generating means for generating a descriptor for the signal according to the category of classification thereof,
• wherein said feature extracting means uses one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block as said characteristic quantities.

A method for retrieving signals comprising:

• dividing an input signal into blocks having a predetermined time length;
• extracting one or more than one characteristic quantities of a signal attribute from the signal of each block;
• classifying the signal of each block into a category according to the characteristic quantities thereof, wherein said signal of each block is classified into any of the categories formed on the basis of types of structures that signals may have and do not depend on the types of signal sources and
• retrieving the signal according to the result of categorical classification or by using a descriptor generated according to the result of categorical classification,
• wherein one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block are used as said characteristic quantities.

The method for retrieving signals according to claim 55, wherein

• said input signal is an audio signal;
• the categories formed on the basis of signal sources for classifying the audio signal of each block include one or more than one of silence, voice, male voice, female voice, music, vocal music, instrumental music, noise, striking sound, environmental sound, sound of hustle and bustle, clapping sound and cheering sound and are used for categorical classification based on the sound sources;
• and
• a signal is retrieved by using the descriptor reflecting or corresponding to the result of said categorical classification based on the sound sources.

An apparatus for retrieving signals comprising:

• a blocking means for dividing an input signal into blocks having a predetermined time length;
• a feature extracting means for extracting one or more than one characteristic quantities of a signal attribute from the signal of each block;
• a categorical classifying means for classifying the signal of each block into a category according to the characteristic quantities thereof, wherein said categorical classifying means classifies said signal of each block into any of the categories formed on the basis of types of structures that signals may have and do not depend on the types of signal sources;
• and
• a signal retrieving means for retrieving the signal according to the result of categorical classification or by using a descriptor generated according to the result of categorical classification,
• wherein said feature extracting means uses one or more than one of the average and variances of the signal power in the block, the average and variances of the power of a band-pass signal of the signal in the block, the average and variances of the spread of the spectrogram of the signal in the block, the average and variances of the pitch frequency of the signal in the block, the average and variances of the degree of harmonic structurization of the signal in the block, the average and variances of the residue signal of linear predictive analysis of the signal in the block and the average and variances of the pitch gain of the residue signal of linear predictive analysis of the signal in the block as said characteristic quantities.

The apparatus for retrieving signals according to claim 56, wherein

• said input signal is an audio signal;
• the categories formed on the basis of signal sources for classifying the audio signal of each block include one or more than one of silence, voice, male voice, female voice, music, vocal music, instrumental music, noise, striking sound, environmental sound, sound of hustle and bustle, clapping sound and cheering sound and are used for categorical classification based on the sound sources;
• and
• said signal retrieving means retrieves a signal by using the descriptor reflecting or corresponding to the result of said categorical classification based on the sound sources.