Amazon Technologies, Inc.

Srikanth Kiran Kotagiri

Charles Benjamin Franklin Waggoner

Yongjun Wu

Yang Yang

Lei Li

Marc Joliveau

Joshua B Barnard

Nicholas James Benson

Stefan Christian Richter

Amarsingh Buckthasingh Winston

Andrew Stuart Glass

Hai Wei

Sitaraman Ganapathy

Deepthi Nandakumar

Rakesh Tangirala

Ronald Diaz

Srinivas Rajagopalan

Techniques are described for adaptive encoding different portions of media content based on content. Characteristics of GOPs of media content can be determined and used to set encoding parameters for the GOs. The GOPs can be encoded such that one GOP is encoded differently than another GOP if they have different characteristics.

Content adaptive encoding

Techniques are described for server-side fragment insertion and delivery of media content. After receiving a request to play back media content, a manifest is generated that includes fragments associated with the requested media content and fragments associated with different media content. A client device requests fragments from a media server according to the manifest, and the media server provides the correct fragment in response to the request.

Server-side fragment insertion and delivery

Techniques enabled by the present disclosure enable media content to be split into sections that can be encoded by multiple video encoders. The disclosed techniques further enable the separately encoded sections to be stitched back together using coordination logic that does not require bi-directional communication between encoding processes.

Split-and-stitch media content encoding

Techniques are described for verifying that an audio file corresponds to an instance of media content. An audio file is divided into a plurality of audio segments, and a digital fingerprint is generated for each of the plurality of audio segments. A digital signature is generated for the audio file by aggregating the digital fingerprints. The generated digital signature and at least one other digital signature corresponding to an instance of media content are processed as inputs to a linear regression machine learning model, to determine a measure of similarity between the generated digital signature and the at least one other digital signature. The linear regression machine learning model can be trained using a supervised learning approach and a set of training data. Embodiments determine whether the audio file corresponds to the instance of media content, based on the measure of similarity.

Machine learning system for matching groups of related media files

Methods and apparatus are described for combining fragments of media content that correspond to multiple quality levels. A particular combination of fragments may be selected for a client device based, at least in part, on feedback received from the client device. In this manner, adaptive bit rate selection can be simulated for a client device that does not support adaptive bit rate selection.

Combining fragments with different encodings

The disclosure is related to allocation of bits in a media stream. In an example, a video stream is segmented into groups of pixels. A determination of a class type is made for individual ones of the groups of pixels. The determination can be based at least in part on semantic similarity of the class type and of a scene represented in the groups of pixels. A further determination occurs for sets of classified data associated with regions of interest (ROIs) according to the determined class type. Masking data associated with the sets of classified data is provided and confidence measures associated with the sets of classified data and the ROIs are determined. Bits are then allocated for groups of pixels based on the masking data and the confidence measures. Thereafter, a bit stream with the bits can be transmitted for playback on a computing device.

Confidence-based encoding

A technology for aspect ratio validation is provided. An object may be detected in a visual media presentation. The detected object may be compared to a first candidate aspect ratio object of a same class of objects as the object. The detected object may also be compared to a second candidate aspect ratio object, which may also be of a same class of objects as the detected object. A determination may be made as to which of the first and second candidate aspect ratio objects the object corresponds.

Aspect ratio validation

Techniques are described caching media content. The caching can provide the same cached media content to multiple requests based on different protocols.

Growing community of inventors

Redmond, WA, United States of America

Srikanth Kiran Kotagiri