Training a device specific acoustic model
Custom acoustic models can be configured by developers by providing audio files with custom recordings. The custom acoustic model is trained by tuning a baseline model using the audio files. Audio files may contain custom noise to apply to clean speech for training. The custom acoustic model is provided as an alternative to a standard acoustic model. A speech recognition system can select an acoustic model for use upon receiving metadata about the device conditions or type. Speech recognition is performed on speech audio using one or more acoustic models. The result can be provided to developers through the user interface, and an error rate can be computed and also provided.
1 . A method of performing speech recognition comprising:
storing a plurality of acoustic models associated with a device;
receiving metadata including information according to which one or more trained acoustic models, of the plurality of acoustic models that are already trained, is selected;
preselecting, based on the metadata, one or more trained acoustic models of the plurality of acoustic models;
receiving speech audio including natural language utterances;
selecting a trained acoustic model from the preselected one or more trained acoustic models, the selected trained acoustic model being trained with environmental features; and
employing the selected trained acoustic model to recognize speech from the natural language utterances included in the received speech audio.
2 . The method of claim 1 , wherein
the plurality of acoustic models are associated with different device conditions, and
the metadata is indicative of a device condition.
3 . The method of claim 2 , wherein
the device conditions include usage conditions of the device.
4 . The method of claim 3 , wherein
usage conditions of the device provide information regarding one or more hardware and software components of the device for receiving the speech audio or for providing audio feedback to a user of the device.
5 . The method of claim 1 , wherein
the received metadata is stored within the device.
6 . The method of claim 1 , wherein
the plurality of acoustic models are associated with different device types, and
the metadata is indicative of a device type.
7 . The method of claim 6 , wherein
the device type identifies at least one of a model number and serial number of the device.
8 . The method of claim 1 , wherein
employing the selected trained acoustic model to recognize speech comprises extraction of phonemes from the received speech audio.
9 . The method of claim 1 , wherein the device is an indoor appliance.
10 . The method of claim 1 , wherein the device is a mobile device.
11 . The method of claim 1 , wherein the plurality of acoustic models are stored, at least in part, in the cloud.
12 . A non-transitory computer readable medium storing code that, if executed by one or more computers, would cause the one or more computers to:
store a plurality of acoustic models associated with a device;
receive metadata including information according to which one or more trained acoustic models, of the plurality of acoustic models that are already trained, is selected;
preselect, based on the metadata, one or more trained acoustic models of the plurality of acoustic models;
receive speech audio including natural language utterances;
select a trained acoustic model from the preselected one or more trained acoustic models, the selected trained acoustic model being trained with environmental features; and
employ the selected trained acoustic model to recognize speech from the natural language utterances included in the received speech audio.
13 . A method of using a platform for configuring device-specific speech recognition, the method comprising:
receiving a selection of a set of at least two acoustic models appropriate for a specific type of a device, the selection of the set of at least two acoustic models being received by an interaction with a graphical user interface provided by a computer system; and
providing received speech audio and metadata to a speech recognition system associated with the platform.
14 . The method of claim 13 further comprising:
providing a custom acoustic model appropriate for the specific type of the device,
wherein the set of selected acoustic models includes the provided custom acoustic model.
15 . The method of claim 13 further comprising:
providing training data for training an acoustic model appropriate to the specific type of the device; and
selecting an acoustic model trained on the provided training data.
16 . The method of claim 13 , wherein
the metadata identifies an acoustic model of the set according to the specific type of the device.
17 . The method of claim 13 , wherein
the metadata identifies a specific device condition, and
a computer system selects an acoustic model from the set of acoustic models in dependence upon the specific device condition.
18 . The method of claim 13 , wherein
the at least two acoustic models recognize speech by extraction of phonemes from the received speech audio.
19 . The method of claim 13 , wherein the interaction is a physical interaction.
20 . A non-transitory computer readable medium storing code that, if executed by one or more computers, would cause the one or more computers to:
detect information useful for selecting an acoustic model from a plurality of acoustic models that are already trained and indicative of a device condition;
receive speech audio;
transmit the detected information and the received speech audio; and
receive information requested by speech in the speech audio,
wherein the detected information is capable of being employed to preselect and select a trained acoustic model from a plurality of acoustic models associated with different device conditions, the trained acoustic model being trained with environmental features, and the detected information being capable of being used to recognize speech from the received speech audio.