An artificial neural network architecture is provided for processing raw audio waveforms to create speaker representations that are used for text-independent speaker verification and recognition. The artificial neural network architecture includes a strided convolution layer, first and second sequentially connected residual blocks, a transformer layer, and a final fully connected (FC) layer. The strided convolution layer is configured to receive raw audio waveforms from a speaker. The first and the second residual blocks both include multiple convolutional and max pooling layers. The transformer layer is configured to aggregate frame level embeddings to an utterance level embedding. The output of the FC layer creates a speaker representation for the speaker whose raw audio waveforms were inputted into the strided convolution layer.
A system and method are provided for detecting the presence of potential commercials in a video data stream. Each of the commercials has an expected time length that is an integer multiple of a first predetermined time length, and has an overall time length that is equal to or less than a second predetermined time length. Transitions in the audio or video of the video data stream are detected and the time of the transitions are recorded. Time differences between one or more successive transitions are detected. Any time differences that are an integer multiple of the first predetermined time length, and that have an overall time length that is equal to or less than the second predetermined time length are identified. The contents of the video data stream associated with the identified time differences are flagged as potential commercials and sent to a content processing platform for further analysis.
G06K 9/00 - Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
3.
AUTOMATED IDENTIFICATION OF PRODUCT OR BRAND-RELATED METADATA CANDIDATES FOR A COMMERCIAL USING PERSISTENCE OF PRODUCT OR BRAND-RELATED TEXT OR OBJECTS IN VIDEO FRAMES OF THE COMMERCIAL
A method and system are provided for assigning metadata candidates to a commercial by performing image analysis on a plurality of the video frames to identify video frames that include one or more of identifiable product-related logos, brand-related logos, product-related text, or brand-related text which appear in the video frames, capturing frame data for such video frames, calculating a persistence metric from the frame data for each of the identifiable product- related logos, brand-related logos, product-related text, or brand-related text, comparing the persistence metric for each of the identifiable product-related logos, brand-related logos, product-related text, or brand-related text to a predetermined threshold value, and assigning to the commercial, the identified products or brands as metadata candidates when the persistence metric exceeds the predetermined threshold value. The persistence metric defines a percentage of frames, or a number of consecutively analyzed frames, that include the one or more of identifiable product-related logos, brand-related logos, product-related text, or brand-related text.
SYSTEM AND METHOD FOR DETECTING REPEATING CONTENT, INCLUDING COMMERCIALS, IN A VIDEO DATA STREAM USING AUDIO-BASED AND VIDEO-BASED AUTOMATED CONTENT RECOGNITION
Methods and apparatus are provided for detecting potential repeating content, such as commercials, in a video data stream by receiving one or more video data streams, parsing each video data stream into a plurality of segments, creating audio fingerprints of each segment, storing the plurality of audio fingerprints in a database, and identifying any audio fingerprints in the received and parsed one or more video data streams that match audio fingerprints in the database that were previously stored from video data streams that were previously received and parsed. Video fingerprints are then created for these same pairs of segments and a similarity analysis is performed. The results of the video fingerprint analysis is used to make a determination of subsequent actions to be taken by a content processing platform that performs recognition processing of the content associated with the segments identified as being potentially repeating content.
G06F 16/783 - Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
5.
SYSTEM AND METHOD FOR DETERMINING TV TUNE-IN ATTRIBUTION
Methods and apparatus are provided for determining a lift metric regarding effectiveness of a digital ad campaign for an audio-visual work on subsequent viewership of the audio-visual work. Viewed content from a universe of monitored AV audio-visual devices is collected. Identifiers of audio-visual devices that received an ad impression for the audio-visual work are also collected. The lift metric may be determined from statistical analysis of this data.
H04N 21/2668 - Creating a channel for a dedicated end-user group, e.g. by inserting targeted commercials into a video stream based on end-user profiles
H04N 21/24 - Monitoring of processes or resources, e.g. monitoring of server load, available bandwidth or upstream requests
6.
SYSTEM AND METHOD FOR REMOVING ERRONEOUSLY IDENTIFIED TV COMMERCIALS DETECTED USING AUTOMATIC CONTENT RECOGNITION
Methods and apparatus are provided for automatically removing erroneously logged commercials from a listing of commercials that are detected in a video data stream by performing automatic content recognition on the video data stream and detecting the identity of each of the commercials played in a commercial break, temporarily logging the identity and start and end time of each detected commercial in a log of played commercials, forming clusters from commercials that overlap in time and have related content, or have significant overlap in time, forming permutations of commercial break timelines from the detected commercials, ranking the timelines based on best fit criteria and selecting the best fit timeline, permanently logging only the commercials in the best fit timeline, and removing the remaining commercials from the temporary log. The remaining logged commercials are presumed to be either erroneously identified commercials or properly identified commercials with erroneous start and end times.
Methods and apparatus are provided for detecting potential repeating content, such as commercials, in a video data stream by receiving one or more video data streams, parsing each video data stream into a plurality of segments, creating a representation of each segment such as a segment fingerprint, storing the plurality of segment representations in a database, and identifying any segment representations in the received and parsed one or more video data streams that match segment representations in the database that were previously stored from video data streams that were previously received and parsed.
H04N 21/4402 - Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to MPEG-4 scene graphs involving reformatting operations of video signals for household redistribution, storage or real-time display
8.
SYSTEM AND METHOD FOR DETECTING UNKNOWN TV COMMERCIALS FROM A LIVE TV STREAM
Unknown potential commercials are detected in a video data stream that contains segments of program type content, and blocks of commercial content. Each block includes a plurality of successive individual commercials. A library of known commercials is maintained in a first database. A video data stream is received in a video processing engine which includes a search engine that is in communication with the first database. The search engine identifies all known commercials in the video data stream and their respective start and end times. The video processing engine identifies all time segments that are sandwiched between the known commercials. The video processing engine filters out as a potential commercial any identified time segments that are significantly longer than the time length of a commercial. The video processing engine designates content of each of the time segments that were not filtered out as being one or more unknown potential commercials.
patents
