Back to results
US 8,798,994 B2Granted
Resource conservative transformation based unsupervised speaker adaptation
AssigneeInternational Business Machines Corporation
InventorsJohn W. Eckhart, Michael Florio, Radek Hampl, Pavel Krbec, Jonathan Palgon
Life of this patent
Estimating the term…
Abstract
The present invention discloses a solution for conserving computing resources when implementing transformation based adaptation techniques. The disclosed solution limits the amount of speech data used by real-time adaptation algorithms to compute a transformation, which results in substantial computational savings. Appreciably, application of a transform is a relatively low memory and computationally cheap process compared to memory and resource requirements for computing the transform to be applied.
Claims20
A method for performing unsupervised speaker adaptations in a resource conservative speaker adaptation system, the method comprising:
- configuring a value N representing a maximum amount of frames of feature vectors extracted from an utterance for generating a transformation for the utterance regardless of utterance length, wherein configuring the value N comprises:
- monitoring the availability of one or more processing resources associated with the system,
- defining resource constraints based on the availability of the one or more processing resources, wherein the resource constraints is an amount of at least one of time, memory, and computer processing cycles available for speaker adaptation purposes, and
- automatically configuring the value N based on the defined resource constraints that are met, wherein monitoring, defining and configuring are performed dynamically to ensure that unsupervised speaker adaptation processes are able to be performed deterministically within the defined resource constraints;
- caching N amount frames of feature vectors extracted from the utterance for unsupervised speaker adaptation purposes;
- creating a transformation from the cache including N amount frames of feature vectors of the utterance;
- and
- applying the created transformation during an unsupervised speaker adaptation process.
5 more elements
A speech enabled computing device comprising:
- an audio transducer configured to receive audio input;
- a central processing unit configured to execute programmatic instructions;
- a data store configured to store digitally encoded information;
- an unsupervised speaker adaptation engine comprising a set of programmatic instructions stored within the data store and configured to be executed by the central processing unit, wherein said speaker adaptation engine is configured to generate real-time transformations for unsupervised speaker adaptation of utterances received through the audio transducer, wherein transformations generated by the speaker adaptation engine utilize at most N amount of frames of feature vectors extracted from the utterances to generate the transformations regardless of a size of the utterances for which the transforms are generated, wherein a value N for the N amount of frames is determined by:
- monitoring the availability of one or more processing resources associated with the device,
- defining resource constraints based on the availability of the one or more processing resources, wherein the resource constraints is an amount of at least one of time, memory, and computer processing cycles available for speaker adaptation purposes, and
- automatically configuring the value N based on the defined resource constraints that are met, wherein monitoring, defining and configuring are performed dynamically to ensure that unsupervised speaker adaptation processes are able to be performed deterministically within the defined resource constraints.
4 more elements
A computer program product for performing unsupervised speaker adaptations in a resource conservative speaker adaptation system, the computer program product comprising:
- a non-transitory computer usable medium having computer usable program code embodied therewith, the computer usable program code comprising:
- computer usable program code for configuring value N representing a maximum amount of frames of feature vectors extracted from an utterance to be used when generating a transformation for the utterance regardless of utterance length, wherein configuring the value N comprises:
- monitoring the availability of one or more processing resources associated with the system,
- defining resource constraints based on the availability of the one or more processing resources, wherein the resource constraints is an amount of at least one of time, memory, and computer processing cycles available for speaker adaptation purposes, and
- automatically configuring the value N based on the defined resource constraints that are met, wherein monitoring, defining and configuring are performed dynamically to ensure that unsupervised speaker adaptation processes are able to be performed deterministically within the defined resource constraints;
- computer usable program code configured to cache N amount frames of feature vectors extracted from the utterance for unsupervised speaker adaptation purposes;
- computer usable program code configured to create an transformation from the cache including N amount frames of feature vectors of the utterance;
- and
- computer usable program code configured to apply the created transformation during an unsupervised speaker adaptation process.
6 more elements
Dependencies are inferred from the claim text.