Custom acoustic models
The technology disclosed relates to performing speech recognition for a plurality of different devices or devices in a plurality of conditions. This includes storing a plurality of acoustic models associated with different devices or device conditions, receiving speech audio including natural language utterances, receiving metadata indicative of a device type or device condition, selecting an acoustic model from the plurality in dependence upon the received metadata, and employing the selected acoustic model to recognize speech from the natural language utterances included in the received speech audio. Each of speech recognition and the storage of acoustic models can be performed locally by devices or on a network-connected server. Also provided is a platform and interface, used by device developers to select, configure, and/or train acoustic models for particular devices and/or conditions.
1. A method of providing a platform for configuring device-specific speech recognition, the method comprising:
receiving custom noise data from a developer;
training a custom acoustic model using the custom noise data and clean speech data:
providing a user interface for developers to select from a set of at least two acoustic models including the trained custom acoustic model;
receiving, from the developer, a selection of the trained custom acoustic model appropriate for a specific type of a device from the set of at least two acoustic models; and
configuring a speech recognition system to perform device-specific speech recognition by implementing the selected acoustic model.
2. The method of claim 1 , further comprising performing the device-specific speech recognition by:
receiving, from the device of the specific type, speech audio including natural language utterances and metadata associated with the received speech audio;
selecting the implemented acoustic model of the set of at least two acoustic models in dependence upon the received metadata; and
using the acoustic model selected in dependence upon the received metadata to recognize speech from the natural language utterances included in the received speech audio.
3. The method of claim 2 , wherein the metadata identifies the implemented acoustic model according to the specific type of the device.
4. The method of claim 2 , wherein the metadata identifies a specific device condition of the device and the speech recognition system selects the implemented acoustic model in dependence upon the specific device condition.
5. The method of claim 1 , further comprising:
receiving a custom acoustic model appropriate for the specific type of the device; and
providing the custom acoustic model within the user interface to be selected as the implemented acoustic model.
6. The method of claim 1 , further comprising:
receiving, from the developer, training data appropriate for the specific type of the device;
training an acoustic model using the received training data; and
providing the trained acoustic model within the user interface to be the implemented acoustic model.
7. A method of a developer using a platform for configuring device-specific speech recognition, the method comprising:
providing custom noise data through a developer interface provided by a computer system;
receiving, for selection through the developer interface, a trained custom acoustic model that has been trained using (i) the custom noise data provided by the developer and (ii) clean speech data;
selecting, through the developer interface and from a set of at least two acoustic models including the trained custom acoustic model, the trained custom acoustic model to implement that is appropriate for a specific type of a device; and
providing speech audio with metadata indicative of the implemented acoustic model to a speech recognition system associated with the platform.
8. The method of claim 7 , further comprising:
providing, to the developer interface, a custom acoustic model appropriate for the specific type of the device,
wherein the set of at least two acoustic models includes the provided custom acoustic model.
9. The method of claim 7 , further comprising:
providing training data for training an acoustic model appropriate to the specific type of the device; and
selecting, through the developer interface, an acoustic model trained on the provided training data to implement.
10. The method of claim 7 , wherein the metadata identifies the implemented acoustic model according to the specific type of the device.
11. The method of claim 7 wherein the metadata identifies a specific device condition and the computer system selects the implemented acoustic model in dependence upon the specific device condition.
12. The method of claim 7 , further comprising using the trained custom acoustic model on a local device of the specific type.
13. The method of claim 7 , further comprising selecting the trained custom acoustic model as the implemented acoustic model for speech recognition performed by the speech recognition system.