Data-driven global boundary optimization

A non-transitory machine-readable storage medium storing instructions to cause a machine to perform a machine-implemented method comprising:

• extracting portions from segment boundary regions of a plurality of speech segments, each segment boundary region based on a corresponding initial unit boundary;
• defining a vector space that represents the portions of the plurality of speech segments to derive feature vectors;
• for each of a plurality of potential unit boundaries within each segment boundary region, determining an average discontinuity based on distances between the feature vectors in the vector space;
• and
• for each segment, selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

An apparatus comprising:

• a processor, and
• means for extracting, using the processor, portions from segment boundary regions of a plurality of speech segments, each segment boundary region based on a corresponding initial unit boundary, wherein at least one portion overlaps across two segments defining the corresponding initial unit boundary;
• means for defining a vector space that represents the portions of the plurality of speech segments to derive feature vectors;
• for each of a plurality of potential unit boundaries within each segment boundary region, means for determining an average discontinuity based on distances between the feature vectors in the vector space;
• and
• for each segment, means for selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

A system comprising:

• a processing unit coupled to a memory through a bus;
• and
• a process executed from the memory by the processing unit to cause the processing unit to:
• extract portions from segment boundary regions of a plurality of speech segments, each segment boundary region based on a corresponding initial unit boundary;
• define a vector space that represents the portions of the plurality of speech segments to derive feature vectors;
• for each of a plurality of potential unit boundaries within each segment boundary region, determine an average discontinuity based on distances between the feature vectors in the vector space;
• and
• for each segment, select a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

A machine-implemented method comprising:

• setting, using a processor, an initial unit boundary for each segment of a plurality of speech segments, each initial unit boundary defining a segment boundary region and a plurality of potential unit boundaries within each segment boundary region;
• computing a vector space representing the plurality of speech segments to derive feature vectors;
• for each segment, determining an average discontinuity over a plurality of concatenations of candidate units defined by the potential unit boundaries based on how far apart the feature vectors are in the vector space;
• for each segment, selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

A non-transitory machine-readable storage medium storing instructions to cause a machine to perform a machine-implemented method comprising:

• setting an initial unit boundary for each segment of a plurality of speech segments, each initial unit boundary defining a segment boundary region and a plurality of potential unit boundaries within each segment boundary region;
• computing a vector space representing the plurality of speech segments to derive feature vectors;
• for each segment, determining an average discontinuity of candidate units defined by the potential unit boundaries based on how far apart the feature vectors are in the vector space;
• for each segment, selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

An apparatus comprising:

• a processor;
• means for setting, using the processor, an initial unit boundary for each segment of a plurality of speech segments, each initial unit boundary defining a segment boundary region and a plurality of potential unit boundaries within each segment boundary region;
• computing a vector space representing the plurality of speech segments to derive feature vectors;
• for each segment, means for determining an average discontinuity of candidate units defined by the potential unit boundaries based on how far apart the feature vectors are in the vector space;
• for each segment, means for selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.

A system comprising:

• a processing unit coupled to a memory through a bus;
• and
• a process executed from the memory by the processing unit to cause the processing unit to:
• set an initial unit boundary for each segment of a plurality of speech segments, each initial unit boundary defining a segment boundary region and a plurality of potential unit boundaries within each segment boundary region;
• compute a vector space representing the plurality of speech segments to derive feature vectors;
• for each segment, determine an average discontinuity of candidate units defined by the potential unit boundaries based on how far apart the feature vectors are in the vector space;
• for each segment, selecting a new unit boundary from the plurality of potential unit boundaries, wherein the new unit boundary is associated with a minimum average discontinuity.