Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media for learning pronunciations from acoustic sequences. One method includes receiving an acoustic sequence, the acoustic sequence comprising a respective acoustic feature representation at each of a plurality of time steps; for each of the time steps processing the acoustic feature representation through each of one or more recurrent neural network layers to generate a recurrent output; processing the recurrent output for the time step using a phoneme output layer to generate a phoneme representation for the acoustic feature representation for the time step; and processing the recurrent output for the time step using a grapheme output layer to generate a grapheme representation for the acoustic feature representation for the time step; and extracting, from the phoneme and grapheme representations for the acoustic feature representations at each time step, a respective pronunciation for each of one or more words.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for presenting notifications in an enterprise system. In one aspect, a method include actions of obtaining a template that defines (i) trigger criteria for presenting a notification type and (ii) content rules for determining content to include in a notification of the notification type. Additional actions include accessing enterprise resources of an enterprise, the enterprise resources including data describing entities related to the enterprise and relationships among the entities. Further actions include, accessing user information specific to a user and determining that the trigger criteria is satisfied by the enterprise resources and the user information. Additional actions include generating a particular notification of the notification type based at least on the content rules and providing the particular notification to the user.

Abstract: In some examples, a computing device includes at least one processor; and at least one module, operable by the at least one processor to: output, for display at an output device, a graphical keyboard; receive an indication of a gesture detected at a location of a presence-sensitive input device, wherein the location of the presence-sensitive input device corresponds to a location of the output device that outputs the graphical keyboard; determine, based on at least one spatial feature of the gesture that is processed by the computing device using a neural network, at least one character string, wherein the at least one spatial feature indicates at least one physical property of the gesture; and output, for display at the output device, based at least in part on the processing of the at least one spatial feature of the gesture using the neural network, the at least one character string.

Abstract: A method of operating a voice-enabled business directory search system includes receiving category-business pairs, each category-business pair including a business category and a specific business, and establishing a data structure having nodes based on the category-business pairs. Each node of the data structure is associated with one or more business categories and a speech recognition language model for recognizing specific businesses associated with the one or more businesses categories.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving an audio signal and initiating speech recognition tasks by a plurality of speech recognition systems (SRS's). Each SRS is configured to generate a recognition result specifying possible speech included in the audio signal and a confidence value indicating a confidence in a correctness of the speech result. The method also includes completing a portion of the speech recognition tasks including generating one or more recognition results and one or more confidence values for the one or more recognition results, determining whether the one or more confidence values meets a confidence threshold, aborting a remaining portion of the speech recognition tasks for SRS's that have not generated a recognition result, and outputting a final recognition result based on at least one of the generated one or more speech results.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving an audio signal and initiating speech recognition tasks by a plurality of speech recognition systems (SRS's). Each SRS is configured to generate a recognition result specifying possible speech included in the audio signal and a confidence value indicating a confidence in a correctness of the speech result. The method also includes completing a portion of the speech recognition tasks including generating one or more recognition results and one or more confidence values for the one or more recognition results, determining whether the one or more confidence values meets a confidence threshold, aborting a remaining portion of the speech recognition tasks for SRS's that have not generated a recognition result, and outputting a final recognition result based on at least one of the generated one or more speech results.

Abstract: Speech recognition systems may perform the following operations: receiving audio at a computing device; identifying a language associated with the audio; recognizing the audio using recognition models for different versions of the language to produce recognition candidates for the audio, where the recognition candidates are associated with corresponding information; comparing the information of the recognition candidates to identify agreement between at least two of the recognition models; selecting a recognition candidate based on information of the recognition candidate and agreement between the at least two of the recognition models; and outputting data corresponding to the selected recognition candidate as a recognized version of the audio.

Abstract: In one implementation a computer-implemented method includes generating a group of telephone contacts for a first user, wherein the generating includes identifying a second user as a contact of the first user based upon a determination that the second user has at least a threshold email-based association with the first user; and adding the identified second user to the group of telephone contacts for the first user. The method further includes receiving a first request to connect a first telephone device associated with the first user to a second telephone device associated with the second user. The method also includes identifying a contact identifier of the second telephone device using the generated group of telephone contacts for the first user, and initiating a connection between the first telephone device and the second telephone device using the identified contact identifier.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating representations of input sequences. One of the methods includes receiving a grapheme sequence, the grapheme sequence comprising a plurality of graphemes arranged according to an input order; processing the sequence of graphemes using a long short-term memory (LSTM) neural network to generate an initial phoneme sequence from the grapheme sequence, the initial phoneme sequence comprising a plurality of phonemes arranged according to an output order; and generating a phoneme representation of the grapheme sequence from the initial phoneme sequence generated by the LSTM neural network, wherein generating the phoneme representation comprises removing, from the initial phoneme sequence, phonemes in one or more positions in the output order.

Abstract: In one implementation a computer-implemented method includes generating a group of telephone contacts for a first user, wherein the generating includes identifying a second user as a contact of the first user based upon a determination that the second user has at least a threshold email-based association with the first user; and adding the identified second user to the group of telephone contacts for the first user. The method further includes receiving a first request to connect a first telephone device associated with the first user to a second telephone device associated with the second user. The method also includes identifying a contact identifier of the second telephone device using the generated group of telephone contacts for the first user, and initiating a connection between the first telephone device and the second telephone device using the identified contact identifier.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for recognizing speech using neural networks. One of the methods includes receiving an audio input; processing the audio input using an acoustic model to generate a respective phoneme score for each of a plurality of phoneme labels; processing one or more of the phoneme scores using an inverse pronunciation model to generate a respective grapheme score for each of a plurality of grapheme labels; and processing one or more of the grapheme scores using a language model to generate a respective text label score for each of a plurality of text labels.

Abstract: Speech recognition systems may perform the following operations: receiving audio; recognizing the audio using language models for different languages to produce recognition candidates for the audio, where the recognition candidates are associated with corresponding recognition scores; identifying a candidate language for the audio; selecting a recognition candidate based on the recognition scores and the candidate language; and outputting data corresponding to the selected recognition candidate as a recognized version of the audio.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for pruning a language model are disclosed. The methods, systems, and apparatus include actions of selecting a candidate portion of the language model to evaluate for pruning, obtaining an entropy score representing information loss that would result from pruning the candidate portion of the language model, obtaining an acoustic score representing acoustic confusability of one or more words modeled by the candidate portion of the language model, and evaluating whether to prune the candidate portion of the language model using the entropy score and the acoustic score.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, including selecting terms; obtaining an expected phonetic transcription of an idealized native speaker of a natural language speaking the terms; receiving audio data corresponding to a particular user speaking the terms in the natural language; obtaining, based on the audio data, an actual phonetic transcription of the particular user speaking the terms in the natural language; aligning the expected phonetic transcription of the idealized native speaker of the natural language with the actual phonetic transcription of the particular user; identifying, based on the aligning, a portion of the expected phonetic transcription that is different than a corresponding portion of the actual phonetic transcription; and based on identifying the portion of the expected phonetic transcription, designating the expected phonetic transcription as a substitute pronunciation for the corresponding portion of the actual phonetic tr

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for verifying pronunciations. In one aspect, a method includes obtaining a first transcription for an utterance. A second transcription for the utterance is obtained. The second transcription is different from the first transcription. One or more feature scores are determined based on the first transcription and the second transcription. The one or more feature scores are input to a trained classifier. An output of the classifier is received. The output indicates which of the first transcription and the second transcription is more likely to be a correct transcription of the utterance.

Abstract: Aspects of this disclosure are directed to accurately transforming speech data into one or more word strings that represent the speech data. A speech recognition device may receive the speech data from a user device and an indication of the user device. The speech recognition device may execute a speech recognition algorithm using one or more user and acoustic condition specific transforms that are specific to the user device and an acoustic condition of the speech data. The execution of the speech recognition algorithm may transform the speech data into one or more word strings that represent the speech data. The speech recognition device may estimate which one of the one or more word strings more accurately represents the received speech data.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for discovery of problematic pronunciations for automatic speech recognition systems. One of the methods includes determining a frequency of occurrences of one or more n-grams in transcribed text and a frequency of occurrences of the n-grams in typed text and classifying a system pronunciation of a word included in the n-grams as correct or incorrect based on the frequencies. The n-grams may comprise one or more words and at least one of the words is classified as incorrect based on the frequencies. The frequencies of the specific n-grams may be determined across a domain using one or more n-grams that typically appear adjacent to the specific n-grams.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining pronunciations for particular terms. The methods, systems, and apparatus include actions of obtaining audio samples of speech corresponding to a particular term and obtaining candidate pronunciations for the particular term. Further actions include generating, for each candidate pronunciation for the particular term and audio sample of speech corresponding to the particular term, a score reflecting a level of similarity between of the candidate pronunciation and the audio sample. Additional actions include aggregating the scores for each candidate pronunciation and adding one or more candidate pronunciations for the particular term to a pronunciation lexicon based on the aggregated scores for the candidate pronunciations.

Abstract: A method of operating a voice-enabled business directory search system includes receiving category-business pairs, each category-business pair including a business category and a specific business, and establishing a data structure having nodes based on the category-business pairs. Each node of the data structure is associated with one or more business categories and a speech recognition language model for recognizing specific businesses associated with the one or more businesses categories.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for transcribing utterances into written text are disclosed. The methods, systems, and apparatus include actions of obtaining a lexicon model that maps phones to spoken text and obtaining a language model that assigns probabilities to written text. Further includes generating a transducer that maps the written text to the spoken text, the transducer mapping multiple items of the written text to an item of the spoken text. Additionally, the actions include constructing a decoding network for transcribing utterances into written text, by composing the lexicon model, the inverse of the transducer, and the language model.