Abstract: A computer system determines one or more media placement opportunities for a media application at a client device and generates a request message that includes an indication of each media placement opportunity. The request message has a hierarchical framework with a plurality of levels that includes, for each media placement opportunity: (i) a media application context level; and (ii) a sub-context level. The computer system transmits the request message to a server system; and receives, from the server system, a response to the request message that includes one or more media content items selected based on the information included in the hierarchical framework of the request message. The computer system provides, to the media application at the client device, at least one of the one or more media content items within at least one of the media placement opportunities.

Abstract: A method for stabilizing a magnetic field of an inertial measurement unit (IMU), is provided that includes initializing accelerometer and gyroscope (AG) heading data for the IMU and initializing accelerometer, gyroscope and magnetometer (AGM) heading data for the IMU. Determining whether a tracking state exists and completing processing of the AG heading data and the AGM heading data if the tracking state does not exist. Calculating a magnetic field error if the tracking state exists.

Abstract: A computer system associated with a media-providing service obtains a set of text descriptors from a plurality of media items in a pool of media items personalized for a first user. The system selects, from the set of text descriptors, a first text descriptor using first criteria and a second text descriptor using second criteria. The system causes an application associated with the media-providing service, executing on a client device of the first user, to concurrently display a first option for playing back a first playlist of media items associated with the first text descriptor and a second option for playing back a second playlist of media items associated with the second text descriptor. In response to receiving selection by the first user of the first option, the system provides, to the application, media items from the first playlist of media items associated with the first text descriptor.

Abstract: Systems and methods for snapping to seek positions in media content using seek guides are provided. Seek guides may be associated with a media content item. Each seek guide may have a position and a radius. When a user seeks to a reference position, a seek guide selector may select a seek guide to use in setting a new playback position. The seek guide selector may use a probabilistic distribution positioned based on the reference position to select a seek guide. Probabilities may be calculated for each seek guide using a cumulative distribution function, and a seek guide may be selected based on the calculated probabilities.

Abstract: A system for processing audio data, comprising a diffusion data filter coupled to a source of digital audio data, the diffusion data filter configured to generate filtered audio data from the digital audio data. A delay coupled to the source of digital audio data, the delay configured to delay the digital audio data by a predetermined amount. A first multiplier configured to multiply the filtered audio data by a distance gain parameter to generate a first intermediate output. A second multiplier configured to multiply the delayed digital audio data by a complementary distance gain parameter to generate a second intermediate output. An adder configured to add the first intermediate output and the second intermediate output to generate an audio output.

Abstract: An electronic device provides, to a user, a user-curated playlist, the user-curated playlist including an ordered set of media items that were added by the user. While providing a first media item in the ordered set of media items, the electronic device receives a first user input selecting an option to include recommended media items in the user-curated playlist. In response to the first user input, the electronic device updates the user-curated playlist to include a first recommended media item, the first recommended media item selected without user intervention based at least in part on attributes of the user-curated playlist. The first recommended media item is positioned in the user-curated playlist in between media items that were added to the ordered set of media items by the user.

Abstract: A parse arbitrator receives a rule-based parser predictive test result indicating whether a first <intent, slots> set contains at least one predictive slot and a descriptive classifier test result indicating whether the digitized descriptive query is descriptive. The parse arbitrator instructs a fulfillment system to perform a fulfillment operation based on the first <intent, slots> set when the rule-based parser predictive test result and the descriptive classifier test result align and instructs a fulfillment system to perform a fulfillment operation based on a second <intent, slots> set when the rule-based parser predictive test result and the descriptive classifier test result do not align. The first <intent, slots> set has been generated by using a first parser to parse a digitized descriptive query and the second <intent, slots> set has been generated by using a second parser to parse the digitized descriptive query.

Abstract: A method for training a speech synthesis model adapted to output speech in response to input text is provided. The method includes receiving training data for training said speech synthesis model, the training data comprising speech that corresponds to known text. The method includes training said speech synthesis model. The method includes testing said speech synthesis model using a plurality of text sequences. The method includes calculating at least one metric indicating the performance of the model when synthesising each text sequence. The method includes determining from said metric whether the speech synthesis model requires further training. The method includes determining targeted training text from said calculated metrics, wherein said targeting training text is text related to text sequences where the metric indicated that the model required further training.

Abstract: This disclosure is directed to an enhanced audio file generator. One aspect is a method of enhancing input speech in an input audio file, the method comprising receiving the input audio file representing the input speech, wherein the input audio file is recorded at an audio recording device, and generating an enhanced audio file by applying an audio transformation model to the input audio file, wherein applying the audio transformation model to generate the enhanced audio file comprises extracting parameters defining audio features from the input audio file, the parameters including a noise parameter defining noise in the input audio file and one or more other preset parameters respectively defining other audio features, synthesizing clean speech based on the extracted parameters including the noise parameter, wherein synthesizing the clean speech comprises transforming the noise parameter to defined value(s); and generating the enhanced audio file with the synthesized clean speech.

Abstract: A method includes receiving text and inputting the received text in a prediction network. The method further includes generating, using the prediction network, speech data. The prediction network comprises a neural network that is trained to generate expressive speech data from text. The neural network is trained by: receiving a first training dataset comprising audio data and corresponding text data; acquiring a respective expressivity score for each audio sample of the audio data; selecting, from the first training dataset, a first subset of training data based on the respective expressivity scores of the audio data in the first training dataset; generating, for the first subset of training data, prediction audio data for the corresponding text data; and comparing the prediction audio data to the audio data of the first subset of training data.

Abstract: An electronic device identifies a set of seed content items that correspond to violative content items. The electronic device determines, using playback histories indicating consumption of respective content items, connections between a respective content item and a first audience that has consumed the respective content item and that has consumed at least a threshold number of seed content items from the set of seed content items. The electronic device provides information corresponding to the connections as an input to a machine learning model. The electronic device receives, as an output from the machine learning model, likelihoods that respective content items are violative content items and stores a set of content items, selected using the output from the machine learning model, as candidate content items in accordance with a determination that the content item satisfies likelihood criteria.

Abstract: A method for synthesising speech from text includes receiving text and encoding, by way of an encoder module, the received text. The method further includes determining, by way of an attention module, a context vector from the encoding of the received text, wherein determining the context vector comprises at least one of: applying a threshold function to an attention vector and accumulating the thresholded attention vector, or applying an activation function to the attention vector and accumulating the activated attention vector. The method further includes determining speech data from the context vector.

Abstract: Systems and methods for optimizing selection of media content for long-term outcomes are provided. Observational data including intermediate outcomes from an observation period is combined with historical data to select media content based on estimated long-term outcomes at the end of an optimization period. As time passes, the observational data is updated with more intermediate outcomes, allowing more accurate estimates of long-term outcomes to be made. In an example, a predictive model trained using the historical data uses the observational data to estimate distributions of long-term outcomes. An action selector selects samples from the distributions and selects media content based on the samples.

Abstract: Systems, methods, and devices for training and testing utterance based frameworks are disclosed. The training and testing can be conducting using synthetic utterance samples in addition to natural utterance samples. The synthetic utterance samples can be generated based on a vector space representation of natural utterances. In one method, a synthetic weight vector associated with a vector space is generated. An average representation of the vector space is added to the synthetic weight vector to form a synthetic feature vector. The synthetic feature vector is used to generate a synthetic voice sample. The synthetic voice sample is provided to the utterance-based framework as at least one of a testing or training sample.

Abstract: Utterance-based user interfaces can include activation trigger processing techniques for detecting activation triggers and causing execution of certain commands associated with particular command pattern activation triggers without waiting for output from a separate speech processing engine. The activation trigger processing techniques can also detect speech analysis patterns and selectively activate a speech processing engine.

Abstract: An electronic device provides, to a user, a user-curated playlist, the user-curated playlist including an ordered set of media items that were added by the user. While providing a first media item in the ordered set of media items, the electronic device receives a first user input selecting an option to include recommended media items in the user-curated playlist. In response to the first user input, the electronic device updates the user-curated playlist to include a first recommended media item, the first recommended media item selected without user intervention based at least in part on attributes of the user-curated playlist. The first recommended media item is positioned in the user-curated playlist in between media items that were added to the ordered set of media items by the user.

Abstract: Methods, systems, and related products that provide detection of media content items that are under-locatable by machine voice-driven retrieval of uttered requests for retrieval of the media items. For a given media item, a resolvability value and/or an utterance resolve frequency is calculated by a number of playbacks of the media item by a speech retrieval modality to a total number of playbacks of the media item regardless of retrieval modality. In some examples, the methods, systems and related products also provide for improvement in the locatability of an under-locatable media item by collecting and/or generating one or more pronunciation aliases for the under-locatable item.

Abstract: An electronic device obtains a plurality of collections of media content items, each collection of media content items being associated with text. Based on how frequently a first media content item co-occurs with a first descriptor in text for respective collections of media items that include the first media content item, the electronic device generates, without user input, a new collection of media content items for a first user. The new collection of media content items corresponds to the first descriptor and includes the first media content item. The electronic device presents the new collection of media content items to the first user as a recommendation.