Abstract: A set of benchmark text strings may be classified to provide a set of benchmark classifications. The benchmark text strings in the set may correspond to a benchmark corpus of benchmark utterances in a particular language. A benchmark classification distribution of the set of benchmark classifications may be determined. A respective classification for each text string in a corpus of text strings may also be determined. Text strings from the corpus of text strings may be sampled to form a training corpus of training text strings such that the classifications of the training text strings have a training text string classification distribution that is based on the benchmark classification distribution. The training corpus of training text strings may be used to train an automatic speech recognition (ASR) system.

Abstract: Methods and systems for automatic speech recognition and methods and systems for training acoustic language models are disclosed. In accordance with one automatic speech recognition method, an acoustic input data set is analyzed to identify portions of the input data set that conform to a general language and to identify portions of the input data set that conform to at least one dialect of the general language. In addition, a general language model and at least one dialect language model is applied to the input data set to perform speech recognition by dynamically selecting between the models in accordance with each of the identified portions. Further, speech recognition results obtained in accordance with the application of the models is output.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, relating to enhanced maximum entropy models. In some implementations, data indicating a candidate transcription for an utterance and a particular context for the utterance are received. A maximum entropy language model is obtained. Feature values are determined for n-gram features and backoff features of the maximum entropy language model. The feature values are input to the maximum entropy language model, and an output is received from the maximum entropy language model. A transcription for the utterance is selected from among a plurality of candidate transcriptions based on the output from the maximum entropy language model. The selected transcription is provided to a client device.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving audio data that corresponds to an utterance, obtaining a first transcription of the utterance that was generated using a limited speech recognizer. The limited speech recognizer includes a speech recognizer that includes a language model that is trained over a limited speech recognition vocabulary that includes one or more terms from a voice command grammar, but that includes fewer than all terms of an expanded grammar. A second transcription of the utterance is obtained that was generated using an expanded speech recognizer. The expanded speech recognizer includes a speech recognizer that includes a language model that is trained over an expanded speech recognition vocabulary that includes all of the terms of the expanded grammar. The utterance is classified based at least on a portion of the first transcription or the second transcription.

Abstract: A method includes accessing data specifying a set of actions, each action defining a user device operation and for each action: accessing a corresponding set of command sentences for the action, determining first n-grams in the set of command sentences that are semantically relevant for the action, determining second n-grams in the set of command sentences that are semantically irrelevant for the action, generating a training set of command sentences from the corresponding set of command sentences, the generating the training set of command sentences including removing each second n-gram from each sentence in the corresponding set of command sentences for the action, and generating a command model from the training set of command sentences configured to generate an action score for the action for an input sentence based on: first n-grams for the action, and second n-grams for the action that are also second n-grams for all other actions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for language models using non-linguistic context. In some implementations, context data indicating non-linguistic context for the utterance is received. Based on the context data, feature scores for one or more non-linguistic features are generated. The feature scores for the non-linguistic features are provided to a language model trained to process scores for non-linguistic features. The output from the language model is received, and a transcription for the utterance is determined using the output of the language model.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving audio data that corresponds to an utterance, obtaining a first transcription of the utterance that was generated using a limited speech recognizer. The limited speech recognizer includes a speech recognizer that includes a language model that is trained over a limited speech recognition vocabulary that includes one or more terms from a voice command grammar, but that includes fewer than all terms of an expanded grammar. A second transcription of the utterance is obtained that was generated using an expanded speech recognizer. The expanded speech recognizer includes a speech recognizer that includes a language model that is trained over an expanded speech recognition vocabulary that includes all of the terms of the expanded grammar. The utterance is classified based at least on a portion of the first transcription or the second transcription.

Abstract: A method iteratively processes data for a set of actions, including: for each action: accessing a corresponding set of command sentences for the action, determining first n-grams that are semantically relevant for the action and second n-grams that are semantically irrelevant for the action, and identifying, from a log of command sentences that includes command sentences not included in the corresponding set of command sentences, candidate command sentences that include one first n-gram and a third n-gram that has not yet been determined to be a first n-gram or a second n-gram; for each candidate command sentence, determining each third n-gram that is semantically relevant for an action to be a first n-gram, and determining each third n-gram that is semantically irrelevant for an action to be a second n-gram, and adjusting the corresponding set of command sentences for each action based on the first n-grams and the second n-grams.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for semantic parsing with multiple parsers. One of the methods includes obtaining one or more transcribed prompt n-grams from a speech to text recognizer, providing the transcribed prompt n-grams to a first semantic parser that executes on the user device and accesses a first knowledge base for results responsive to the spoken prompt, providing the transcribed prompt n-grams to a second semantic parser that accesses a second knowledge base for results responsive to the spoken prompt, the first knowledge base including first data not included in the second knowledge base, receiving a result responsive to the spoken prompt from the first semantic parser or the second semantic parser, wherein the result is selected from the knowledge base associated with the semantic parser that provided the result to the user device, and performing an operation based on the result.

Abstract: A method includes accessing data specifying a set of actions, each action defining a user device operation and for each action: accessing a corresponding set of command sentences for the action, determining first n-grams in the set of command sentences that are semantically relevant for the action, determining second n-grams in the set of command sentences that are semantically irrelevant for the action, generating a training set of command sentences from the corresponding set of command sentences, the generating the training set of command sentences including removing each second n-gram from each sentence in the corresponding set of command sentences for the action, and generating a command model from the training set of command sentences configured to generate an action score for the action for an input sentence based on: first n-grams for the action, and second n-grams for the action that are also second n-grams for all other actions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating expressions associated with voice commands. The methods, systems, and apparatus include actions of obtaining segments of one or more expressions associated with a voice command. Further actions include combining the segments into a candidate expression and scoring the candidate expression using a text corpus. Additional actions include selecting the candidate expression as an expression associated with the voice command based on the scoring of the candidate expression.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving audio data that corresponds to an utterance, obtaining a first transcription of the utterance that was generated using a limited speech recognizer. The limited speech recognizer includes a speech recognizer that includes a language model that is trained over a limited speech recognition vocabulary that includes one or more terms from a voice command grammar, but that includes fewer than all terms of an expanded grammar. A second transcription of the utterance is obtained that was generated using an expanded speech recognizer. The expanded speech recognizer includes a speech recognizer that includes a language model that is trained over an expanded speech recognition vocabulary that includes all of the terms of the expanded grammar. The utterance is classified based at least on a portion of the first transcription or the second transcription.

Abstract: In some implementations, audio data for an utterance is provided over a network. At a client device and over the network, information is received that indicates candidate transcriptions for the utterance and semantic information for the candidate transcriptions. A semantic parser is used at the client device to evaluate each of at least a plurality of the candidate transcriptions. One of the candidate transcriptions is selected based on at least the received semantic information and the output of the semantic parser for the plurality of candidate transcriptions that are evaluated.

Abstract: A computer-implemented method can include receiving, by a computer system, a request to transcribe spoken input from a user of a computing device, the request including information that (i) characterizes a spoken input, and (ii) context information associated with the user or the computing device. The method can determine, based on the information that characterizes the spoken input, multiple hypotheses that each represent a possible textual transcription of the spoken input. The method can select, based on the context information, one or more of the multiple hypotheses for the spoken input as one or more likely intended hypotheses for the spoken input, and can send the one or more likely intended hypotheses for the spoken input to the computing device. In conjunction with sending the one or more likely intended hypotheses for the spoken input to the computing device, the method can delete the context information.

Abstract: Method for online character recognition of Arabic text, the method including receiving handwritten Arabic text from a user in the form of handwriting strokes, sampling the handwriting strokes to acquire a sequence of two dimensional point representations thereof, with associated temporal data, geometrically pre processing and extracting features on the point representations, detecting delayed strokes and word parts in the pre processed point representations, projecting the delayed strokes onto the body of the word parts, constructing feature vector representations for each word part, thereby generating an observation sequence, and determining the word with maximum probability given the observation sequence, resulting in a list of word probabilities.

Abstract: Methods and systems for automatic speech recognition and methods and systems for training acoustic language models are disclosed. One system for automatic speech recognition includes a dialect recognition unit and a controller. The dialect recognition unit is configured to analyze acoustic input data to identify portions of the acoustic input data that conform to a general language and to identify portions of the acoustic input data that conform to at least one dialect of the general language. In addition, the controller is configured to apply a general language model and at least one dialect language model to the input data to perform speech recognition by dynamically selecting between the models in accordance with each of the identified portions.

Abstract: A set of benchmark text strings may be classified to provide a set of benchmark classifications. The benchmark text strings in the set may correspond to a benchmark corpus of benchmark utterances in a particular language. A benchmark classification distribution of the set of benchmark classifications may be determined. A respective classification for each text string in a corpus of text strings may also be determined. Text strings from the corpus of text strings may be sampled to form a training corpus of training text strings such that the classifications of the training text strings have a training text string classification distribution that is based on the benchmark classification distribution. The training corpus of training text strings may be used to train an automatic speech recognition (ASR) system.

Abstract: Disclosed embodiments relate to mapping an utterance to an action using a classifier. One illustrative computing device includes a user interface having an input component. The computing device further includes a processor and a computer-readable storage medium, having stored thereon program instructions that, upon execution by the processor, cause the computing device to perform a set of operations including: receiving an audio utterance via the input component; determining a text string based on the utterance; determining a string-feature vector based on the text string; selecting a target classifier from a set of classifiers, wherein the target classifier is selected based on a determination that a string-feature criteria of the target classifier corresponds to at least one string-feature of the string-feature vector; and initiating a target action that corresponds to the target classifier.

Abstract: Methods and systems for log mining for grammar-based text processing are provided. A method may comprise receiving, from a device, an activity log. The activity log may comprise one or more of an input instruction, a determined function based at least in part on a match of the input instruction to a grammar-based textual pattern including associations of a given function based on one or more grammars, and a response determination based on an acknowledgement of the determined function. The method may also comprise comparing at least a portion of the activity log with stored activity logs in order to determine a correlation between the activity log and the stored activity logs. The method may also comprise modifying the grammar-based textual pattern based on the determined correlation and providing information indicative of the modification to the device so as to update the grammar-based textual pattern.

Abstract: A set of benchmark text strings may be classified to provide a set of benchmark classifications. The benchmark text strings in the set may correspond to a benchmark corpus of benchmark utterances in a particular language. A benchmark classification distribution of the set of benchmark classifications may be determined. A respective classification for each text string in a corpus of text strings may also be determined. Text strings from the corpus of text strings may be sampled to form a training corpus of training text strings such that the classifications of the training text strings have a training text string classification distribution that is based on the benchmark classification distribution. The training corpus of training text strings may be used to train an automatic speech recognition (ASR) system.