Spotify Ab

Rachel Malia Bittner

Tristan Jehan

Andreas Jansson

Angus William Sackfield

Ching Chuan Sung

Andreas Simon Thore Jansson

Nicola Montecchio

Minwei Gu

Sravana Reddy

Lu Han

Eric J Humphrey

Hunter McCurry

Gandalf Hernandez

Mateo Rando

Miles Lennon

Ulf Karlsson

Thor Kell

Michael Scibor

Frej Connoly

A system, method and computer product for training a neural network system. The method comprises inputting an audio signal to the system to generate plural outputs f(X, Θ). The audio signal includes one or more of vocal content and/or musical instrument content, and each output f(X, Θ) corresponds to a respective one of the different content types. The method also comprises comparing individual outputs f(X, Θ) of the neural network system to corresponding target signals. For each compared output f(X, Θ), at least one parameter of the system is adjusted to reduce a result of the comparing performed for the output f(X, Θ), to train the system to estimate the different content types. In one example embodiment, the system comprises a U-Net architecture. After training, the system can estimate various different types of vocal and/or instrument components of an audio signal, depending on which type of component(s) the system is trained to estimate.

Automatic isolation of multiple instruments from musical mixtures

Systems, devices, apparatuses, components, methods, and techniques for automatically generating media previews are provided. An example media system for automatically generating media previews for a particular artist include a trailer generation application configured to receive input specifying an artist and duration of a trailer, automatically select clips from two or more media items by the artist, and automatically arrange and combine the clips into a media trailer for later playback.

Automatically generated media preview

A call to action processor receives an entity datapoint containing data related to an entity, a campaign objective datapoint containing data associated with a campaign objective, at least one definite script element based on the campaign objective, and entity metadata containing data associated with the entity. The call to action further performs generating at least one variable script element based on the entity metadata, presenting to a device the at least one definite script element the at least one variable script element.

Systems, methods and computer program products for generating script elements and call to action components therefor

An electronic device receives a first audio content item that includes a plurality of sound sources. The electronic device generates a representation of the first audio content item. The electronic device determines, from the representation of the first audio content item: a representation of an isolated sound source, and frequency data associated with the isolated sound source. Determining the representation of the isolated sound source and the frequency data associated with the isolated sound source includes using a neural network to jointly determine the representation of the isolated sound source and the frequency data associated with the isolated sound source. The electronic device determines that a portion of a second audio content item matches the first audio content item using the representation of the isolated sound source and/or the frequency data associated with the isolated sound source.

Systems and methods for jointly estimating sound sources and frequencies from audio

Media content playback is based on an identified geographic location of a sound source. A geolocation of a target venue is determined. A first geolocation of a media playback device is determined. An indication that the media playback device is within a threshold distance of the target venue is received. Media content associated with the target venue is transferred to the media playback device, wherein one or more characteristics of the media content are dynamically modified to simulate audio originating from the target device. The one or more characteristics of the media content can include audio content characteristics associated with a timing filter, a loudness filter, and a timbre filter.

Media content playback based on an identified geolocation of a target venue

A cuepoint determination system utilizes a convolutional neural network (CNN) to determine cuepoint placements within media content items to facilitate smooth transitions between them. For example, audio content from a media content item is normalized to a plurality of beats, the beats are partitioned into temporal sections, and acoustic feature groups are extracted from each beat in one or more of the temporal sections. The acoustic feature groups include at least downbeat confidence, position in bar, peak loudness, timbre and pitch. The extracted acoustic feature groups for each beat are provided as input to the CNN on a per temporal section basis to predict whether a beat immediately following the temporal section within the media content item is a candidate for cuepoint placement. A cuepoint placement is then determined from among the candidate cuepoint placements predicted by the CNN.

Cuepoint determination system

Systems, devices, apparatuses, components, methods, and techniques for providing media content to a media playback system are provided. The techniques provide for determining whether a playback system is connected, or otherwise associated with, a playback device suitable for 3D audio playback. Upon determining that a playback system is or is not connected or otherwise associated with a playback device suitable for 3D audio playback, audio content in a corresponding format may then be transferred and/or played by a requesting system. In some examples, one or more filters may be applied to requested audio content to compensate for determined user head movement and/or to create simulated 3D audio from generic two-channel recording.

Audio content format selection

Term masking is performed by generating a time-alignment value for a plurality of units of sound in vocal audio content contained in a mixed audio track, force-aligning each of the plurality of units of sound to the vocal audio content based on the time-alignment value, thereby generating a plurality of force-aligned identifiable units of sound, identifying from the plurality of force-aligned units of sound a force-aligned unit of sound to be altered, and altering the identified force-aligned unit of sound.

Masking systems and methods

A system, method and computer product for training a neural network system. The method comprises inputting an audio signal to the system to generate plural outputs f(X, &#x398;). The audio signal includes one or more of vocal content and/or musical instrument content, and each output f(X, &#x398;) corresponds to a respective one of the different content types. The method also comprises comparing individual outputs f(X, &#x398;) of the neural network system to corresponding target signals. For each compared output f(X, &#x398;), at least one parameter of the system is adjusted to reduce a result of the comparing performed for the output f(X, &#x398;), to train the system to estimate the different content types. In one example embodiment, the system comprises a U-Net architecture. After training, the system can estimate various different types of vocal and/or instrument components of an audio signal, depending on which type of component(s) the system is trained to estimate.

A system of playing media content items determines transitions between pairs of media content items by determining desirable locations in which transitions across the pairs of media content items occur. The system uses a plurality of track features of media content items and determines such track features of each media content item associated with each of transition point candidates, such as beat positions, of that media content item. The system determines similarity in the plurality of track features between the transition point candidates of a first media content item and the transition point candidates for a second media content item being played subsequent to the first media content item. The transition points or portions of the first and second media content items are selected from the transition point candidates for the first and second media content items based on the similarity results.

Transitions between media content items

An electronic device receives a first audio content item that includes a plurality of sound sources. The electronic device generates a representation of the first audio content item. The electronic device determines, from the representation of the first audio content item, a representation of an isolated sound source and frequency data associated with the isolated sound source. The determining includes using a neural network to jointly determine the representation of the isolated sound source and the frequency data associated with the isolated sound source.

Term masking is performed by generating a time-alignment value for a plurality of identifiable units of sound in vocal audio content contained in a mixed audio track, force-aligning each of the plurality of identifiable units of sound to the vocal audio content based on the time-alignment value, thereby generating a plurality of force-aligned identifiable units of sound, identifying from the plurality of force-aligned identifiable units of sound a force-aligned identifiable unit of sound to be muddled, and audio muddling the force-aligned identifiable unit of sound to be muddled.

Growing community of inventors

Brooklyn, NY, United States of America

Rachel Malia Bittner