ImageRepositoryContentBasedVideoMetadata

Timestamp of the last successful Ares classification request.

video_preview_bytes is only exported as virtual dataset by IE VideoUnderstanding and should not be persisted. It will be used by downstream IE functions to push for serving.

When Transcode itag 140 is requested, MediaAnalyzer (as the part of Viper graph) generates audio info including loudness_data, which is then published to Streamer. For Audio news client, we extract this loudness data from Streamer to this field.

Transcript generated from Cloud Speech API

Video Understanding Golden features. (go/amarna-video-signals#golden-signals) Note: Golden6 features (names matching "video_") are DEPRECATED. Please migrate to Golden7 ("VideoFeatures."). For more context, see go/golden7/migrating-from-golden6 and go/amarna-golden-feature-tracker. Signals popluated in Raffia cdoc.doc_videos are configured in cs/symbol:AMARNA_EXPORTED_GOLDEN7_FEATURES.

The video porn confidence score extracted from WhisperV4 featureSet: "VideoFeatures.whisper_v4_labels", with CR2 label: "/tns/porn".

Speech related metadata The transcript_asr field is generated from the YT caption's SPEECH_RECOGNIZER asset.

LMS preview frame perdocs. Timestamps of the frame perdocs are from the original video, not from the preview.

Amarna signals blob that contains large-size signals like VCA frame-level signals.

Frame-level Starburst embeddings. They are IE only signals for short videos initially and will be supported for all videos later. (go/frame-level-sbv5-on-ie)

Information on frame-level people feature blob. (go/vpf-frame-features)

Deprecated, please use media_info.

Metadata about each transcode requested.

Data about whether or not the video was truncated.

Information about the media file, such as duration, resolution, and detail about each audio/video stream. Note that it contains no PII.

The video id in the venom pipeline for STAMP purposes. DEPRECATED: Use transcode_metadata or venom_processing_info instead, which includes the ID and contains information for all clients.

Trnascript generated through AMARNA_CLOUD_SPEECH asset in Venom. Note that AMARNA_CLOUD_SPEECH uses S3 as the speech engine backend, similar to YT caption's SPEECH_RECOGNIZER asset. However, they may use different S3 models.

Information about the video's status in Venom, including IDs and processing times.

Contains SafeSearch video classification outputs which are vertical_name/float pairs.

The video porn confidence score extracted from Whisper featureSet: "video_labels:whisper_v3", with CR2 label: "/cr2/1".

Thumbnail quality score predict how visual pleasing a thumbnail is, based on the model trained with deep neural networks.(go/thumb_features_dd) Note the signal currently only available for Youtube videos.

Set from the video header if truncated, or is the verified length if completely crawled.

Metadata related to Inline playback on the Interest Feed. This field is filled by Hamilton.

Information for the amarna signals blob.

Both audio- and audio-video-files are treated as videos during indexing (whether they share a container format, like .mp4, or not, like .mp3). This bool indicates that there's no video track, just an audio track.

Golden7 video-level people features. (go/ypf-video-features)

Video anchor sets hold set of anchors with multiple anchor types and sequence of VideoAnchor which contains metadata about the anchor, such as thumbnail, perdoc data.

A hash of the video bytes used as a key to Amarna's video_metadata table.

multi_thumbnails_frame_perdocs contains perdocs(go/vss-signals#perdoc) for all frame thumbnails generated by multiple frames per minutes. See more details in go/multiple-thumbnails-per-minutes.

video_previews contain the preview metadata but no bytes. It exits for IE and non-IE cases.

Legos results

Contains lists of reasons why YT videos were filtered from specific processing.

If set, video has been deleted using the deletion service (MediaDeletionService).

64 bit docid used for retrieving video previews.

VideoTranscriptAnnotations holds sentence segmented text and timing information to be used for VideoAnswers (go/video-answers). Note that only punctuated_transcript, timing_info, and lang field are filled, and other fields will be filled in the later stage.