Voice analysis device, voice analysis method and voice analysis program

A computer-executable method of animation image generation which displays an image corresponding to partitions of phonemes from an input voice signal, comprising:

• a first step specifying a time-point in said input voice signal;
• a second step extracting the voice signal contained in a time range of prescribed length from said time-point;
• and
• a third step decomposing said extracted voice signal into frequency component data, wherein
• n frequency component data extracted from the voice signal contained in n time ranges of aid prescribed length are found by repeating said first, second and third steps n times (where n is a natural number of at least 6) for each prescribed time;
• (n−1) correlations are found from n said frequency component data by obtaining i-th correlation (where i is a natural number of at least 1 and no more than (n−1)) based on the i-th item of said frequency component data and the (i+1)-th item of said frequency component data corresponding to said voice signal contained in mutually adjacent time ranges of said prescribed length;
• (n−2) degrees of change are found from the (n−1) said correlations, by finding the k-th degree of change based on the k-th correlation and said (k+1)-th said correlation (where k is a natural number of at least 1 and not greater than (n−2));
• when m is defined as the number specifying the degree of change that is larger than two degrees of change adjacent to the specified degree of change in the (n−2) degrees of change from the first to the (n−2)-th, namely m satisfying the condition that the m-th degree of change is larger than the (m−1)-th degree of change and is larger than the (m+1)-th degree of change when m is changed one at a time from 2 to (n−3), the defined number m is found;
• a time range of said prescribed length is specified in accordance with the defined number m;
• said input voice signal is partitioned into a plurality of sections in accordance with said specified time ranges;
• characteristic quantities are calculated for each section of the partitioned voice signals;
• said phonemes of said sections are specified by comparing said characteristic quantities with reference data for each said phoneme;
• and
• images corresponding to said phonemes of said sections are displayed, being changed over in a manner that is matched with said sections.

A storage medium storing a voice analysis program which is executable by a computer for detecting boundaries of phonemes from input voice, wherein said program when executed by the computer causes the computer to perform:

• a first step of specifying a time-point in said input voice signal;
• a second step of extracting the voice signal contained in a time range of prescribed length from said time-point;
• and
• a third step of decomposing said extracted voice signal into frequency component data, wherein
• n frequency component data extracted from the voice signal contained in n time ranges of said prescribed length are found by repeating said first, second and third steps by said CPU n times (where n is a natural number of at least 6) for each prescribed time;
• (n−1) correlations are found from n said frequency component data by obtaining i-th correlation (where i is a natural number of at least 1 and no more than (n−1)) based on the i-th item of said frequency component data and the (i+1)-th item of said frequency component data corresponding to said voice signal contained in mutually adjacent time ranges of said prescribed length;
• (n−2) degrees of change are found from the (n−1) said correlations, by finding the k-th degree of change based on the k-th correlation and said (k+1)-th said correlation (where k is a natural number of at least 1 and not greater than (n−2));
• when m is defined as the number specifying the degree of change that is larger than two degrees of change adjacent to the specified degree of change in the (n−2) degrees of change from the first to the (n−2)-th, namely m satisfying the condition that the m-th degree of change is larger than the (m−1)-th degree of change and is larger than the (m+1)-th degree of change when m is changed one at a time from 2 to (n−3), the defined number m is found;
• a time range of said prescribed length is specified in accordance with the defined number m;
• and
• said input voice signal is partitioned by said CPU into a plurality of sections in accordance with said specified time ranges.

A storage medium storing an animation image generating program which is executable by a computer for displaying an image corresponding to partitions of phonemes from an input voice signal, wherein said program when executed by said computer causes the computer to perform:

• a first step specifying a time-point in said input voice signal;
• a second step extracting the voice signal contained in a time range of prescribed length from said time-point;
• and
• a third step decomposing said extracted voice signal into frequency component data, wherein
• n frequency component data extracted from the voice signal contained in n time ranges of said prescribed length are found by repeating said first, second and third steps n times (where n is a natural number of at least 6) for each prescribed time;
• (n−1) correlations are found from n said frequency component data by obtaining i-th correlation (where i is a natural number of at least 1 and no more than (n−1)) based on the i-th item of said frequency component data and the (i+1)-th item of said frequency component data corresponding to said voice signal contained in mutually adjacent time ranges of said prescribed length;
• (n−2) degrees of change are found from the (n−1) said correlations, by finding the k-th degree of change based on the k-th correlation and said (k+1)-th said correlation (where k is a natural number of at least 1 and not greater than (n−2));
• when m is defined as the number specifying the degree of change that is larger than two degrees of change adjacent to the specified degree of change in the (n−2) degrees of change from the first to the (n−2)-th, namely m satisfying the condition that the m-th degree of change is larger than the (m−1)-th degree of change and is larger than the (m+1)-th degree of change when m is changed one at a time from 2 to (n−3), the defined number m is found;
• a time range of said prescribed length is specified in accordance with the defined number m;
• said input voice signal is partitioned into a plurality of sections in accordance with said specified time ranges;
• characteristic quantities are calculated for each section of the partitioned voice signals;
• said phonemes of said sections are specified by comparing said characteristic quantities with reference data for each said phoneme;
• and
• images corresponding to said phonemes of said sections are displayed, being changed over in a manner that is matched with said sections.

An animation image generating device comprising:

• a processor to execute a program to perform as:
• a frequency decomposition unit that finds n frequency component data extracted from a voice signal contained in n time ranges of prescribed length by repeating n times (where n is a natural number of at least 6), at prescribed intervals, to specify a time-point in an input voice signal, to extract a voice signal contained in a time range of the prescribed length from said time-point and to decompose of said extracted voice signal into frequency component data;
• a degree of change calculation unit that finds (n−1) correlations from n said frequency component data by obtaining i-th correlation (where i is a natural number of at least 1 and no more than (n−1)) based on the i-th item of said frequency component data and the (i+1)-th item of said frequency component data corresponding to said voice signal contained in mutually adjacent time ranges of said prescribed length, and that finds (n−2) said degrees of change from the (n−1) said correlations by finding the k-th degree of change based on the k-th correlation and said (k+1)-th said correlation (where k is a natural number of at least 1 and not greater than (n−2));
• a segment determining unit that finds a defined number, the defined number m being defined as the number specifying the degree of change that is larger than two degrees of change adjacent to the specified degree of change in the (n−2) degrees of change from the first to the (n−2)-th, namely the defined number m satisfying the condition that the m-th degree of change is larger than the (m−1)-th degree of change and is larger than the (m+1)-th degree of change when m is changed one at a time from 2 to (n−3), that specifies a time range of said prescribed length in accordance with the defined number m, and that partitions said input voice signal into a plurality of sections in accordance with said specified time ranges;
• and
• a phoneme extraction unit that specifies the said phonemes of said sections by calculating characteristic quantities for each section of the partitioned voice signal and comparing reference data of each of said phonemes and said characteristic quantities;
• and
• an animation display unit to display images corresponding to said phonemes of said sections, these images being changed over in a manner that is matched with said sections.

A voice analysis device comprising:

• a processor to execute a program to perform as:
• a frequency decomposition unit that finds n frequency component data extracted from a voice signal contained in n time ranges of prescribed length by repeating n times (where n is a natural number of at least 6), at prescribed intervals, to specify a time-point in an input voice signal, to extract a voice signal contained in a time range of the prescribed length from said time-point and to decompose of said extracted voice signal into frequency component data;
• a degree of change calculation unit that finds (n−1) correlations from n said frequency component data by obtaining i-th correlation (where i is a natural number of at least 1 and no more than (n−1)) based on the i-th item of said frequency component data and the (i+1)-th item of said frequency component data corresponding to said voice signal contained in mutually adjacent time ranges of said prescribed length, and that finds (n−2) said degrees of change from the (n−1) said correlations by finding the k-th degree of change based on the k-th correlation and said (k+1)-th said correlation (where k is a natural number of at least 1 and not greater than (n−2));
• and
• a segment determining unit that finds a defined number, the defined number m being defined as the number specifying the degree of change that is larger than two degrees of change adjacent to the specified degree of change in the (n−2) degrees of change from the first to the (n−2)-th, namely the defined number m satisfying the condition that the m-th degree of change is larger than the (m−1)-th degree of change and is larger than the (m+1)-th degree of change when m is changed one at a time from 2 to (n−3), that specifies a time range of said prescribed length in accordance with the defined number m, and that partitions said input voice signal into a plurality of sections in accordance with said specified time ranges.