Resources

Companion websites for publications

• Seeing Sound: Investigating the Effects of Visualizations and Complexity on Crowdsourced Audio Annotations

Data

• SONYC-UST v2.3 (a.k.a SONYC-UST with Spatiotemporal Context) Dataset – A multilabel (23 class) dataset containing 18515 sound recordings from the SONYC acoustic sensor network that includes metadata on where and when the audio was recorded. All recordings were labeled by Zooniverse volunteers. A verified subset set (1385 recordings) is annotated by the SONYC team.
• SONYC-UST v1.0 Dataset – A multilabel (23 class) dataset containing 2794 sound recordings from the SONYC acoustic sensor network. All recordings were labeled by Zooniverse volunteers. A validation set (443 recordings) is annotated by the SONYC team.
• Urbansound Dataset – A dataset containing 1302 labeled sound recordings. Each recording is labeled with the start and end times of sound events from 10 classes
• Urbansound8k Dataset – A dataset containing 8732 labeled sound excerpts (<=4s) of urban sounds from 10 classes
• URBAN-SED Dataset – A dataset of 10,000 synthesized soundscapes with sound event annotations generated using Scaper 
• Seeing Sound Dataset – A dataset of 5400 crowdsourced audio annotations of 60 synthesized soundscapes

Code

• DCASE 2020 Urban Sound Tagging with Spatiotemporal Context Model – Python Baseline system for the DCASE 2020 task
• DCASE 2019 Urban Sound Tagging Model – Python Baseline system for the DCASE 2019 task
• OpenL3 – A Python library for extracting deep audio embeddings trained on audio-visual correspondence
• AutoPool – an adaptive pooling operator, useful for training audio classification models that produce frame-wise outputs using clip-level training labels (i.e., multiple-instance learning)
• Scaper – A Python library for soundscape synthesis and augmentation
• Audio-Annotator – A Javascript web interface for annotating audio data
• Raster Join
• Urban Pulse