Speech recognition candidate selection based on non-acoustic input
AssigneeInternational Business Machines Corporation
InventorsJonathan H. Connell, II, Etienne Marcheret
Life of this patent
Estimating the term…
Abstract
A method includes the following steps. A speech input is received. At least two speech recognition candidates are generated from the speech input. A scene related to the speech input is observed using one or more non-acoustic sensors. The observed scene is segmented into one or more regions. One or more properties for the one or more regions are computed. One of the speech recognition candidates is selected based on the one or more computed properties of the one or more regions.
Claims20
A method, comprising:
- receiving a speech input;
- generating at least two speech recognition candidates from the speech input;
- observing a scene related to the speech input using one or more non-acoustic sensors;
- segmenting the observed scene into a plurality of regions, wherein each of the regions corresponds to an object or a surface in the observed scene;
- computing properties for at least a given region of the plurality of regions, wherein computing the properties for the given region comprises computing one or more characteristics for the given region and computing one or more relationships between the given region and remaining ones of the plurality of regions, and wherein the one or more characteristics of the given region comprise a color, a shape and a textual label;
- and
- selecting one of the speech recognition candidates based at least in part on the computed properties of the given region.
An apparatus, comprising:
- a memory;
- and
- a processor operatively coupled to the memory and configured to:
- receive a speech input;
- generate at least two speech recognition candidates from the speech input;
- observe a scene related to the speech input using one or more non-acoustic sensors;
- segment the observed scene into a plurality of regions, wherein each of the regions corresponds to an object or a surface in the observed scene;
- compute properties for at least a given region of the plurality of regions, wherein the computation of the properties for the given region comprises a computation of one or more characteristics for the given region a computation of one or more relationships between the given region and remaining ones of the plurality of regions, and wherein the one or more characteristics of the given region comprise a color, a shape and a textual label;
- and
- select one of the speech recognition candidates based at least in part on the computed properties of the given region.
An article of manufacture comprising a computer readable storage medium for storing computer readable program code which, when executed, causes a computer to:
- receive a speech input;
- generate at least two speech recognition candidates from the speech input;
- observe a scene related to the speech input using one or more non-acoustic sensors;
- segment the observed scene into a plurality of regions, wherein each of the regions corresponds to an object or a surface in the observed scene;
- compute properties for at least a given region of the plurality of regions, wherein the computation of the properties for the given region comprises a computation of one or more characteristics for the given region a computation of one or more relationships between the given region and remaining ones of the plurality of regions, and wherein the one or more characteristics of the given region comprise a color, a shape and a textual label;
- and
- select one of the speech recognition candidates based at least in part on the computed properties of the given region.
Dependencies are inferred from the claim text.