This dataset is based on AudioSet’s data available here https://research.google.com/audioset/. It contains video IDs and time bounds for youtube clips, as well as 128-dimensional audio-based features created with VGGish based on these clips. 
You must predict if the sound clip from which audio_embedding originates contains a turkey sound. 

AudioSet's dataset is under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, while their ontology is under a Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license. Our data includes data from both, modified to fit a competition format.

## File descriptions

*   **train.json** - the training set
*   **test.json** - the test set
*   **sample_submission.csv** - a sample submission file in the correct format

## Data fields

*   **vid_id** - YouTube video ID associated with this sample. 
*   **start_time_seconds_youtube_clip** - Where in the YouTube video this audio feature starts.
*   **end_time_seconds_youtube_clip** - Where in the YouTube video this audio feature ends.
*   **audio_embedding** - Extracted frame-level audio feature, embedded down to 128 dimensions per frame using AudioSet’s VGGish tools available here: https://github.com/tensorflow/models/tree/master/research/audioset
*   **is_turkey** -  The target: whether or not the original audio clip contained a turkey. Label is a soft label, based on whether or not AudioSet’s ontology labeled this clip with “Turkey”, and may count turkey calls and other related content as being “turkey”. is_turkey is 1 if the clip contains a turkey sound, and 0 if it does not.