A Dataset and Taxonomy for Urban Sound Research

By Justin Salamon, Christopher Jacoby, Juan Pablo Bello. 2014

1,558 citations recorded by the source corpus.

Subfields: Speech and Hearing, Evaluation & Benchmarks, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Speech and Hearing

In Evaluation & Benchmarks

In Speech & Audio

In Signal Processing

Citations in the corpus

MERLOT RESERVE: Neural Script Knowledge through Vision and Language and Sound (2022, methodology): “In comparison, AudioCLIP [53] (which is supervised on human-annotated labels from AudioSet [42]), performs 16% higher on ESC50, and 6.4% higher on US8K.”
FSD50K: An Open Dataset of Human-Labeled Sound Events (2020, methodology): “As we have seen above, several datasets containing Freesound audio have been widely used by the research community [22, 32, 33, 39, 54, 55], showing its usefulness for dataset creation.”
Moments in Time Dataset: One Million Videos for Event Understanding (2018, background): “[35] released a dataset of urban sounds, and Gemmeke et al.”
Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition (2018, background): “Listening to the World[22] demonstrates how combining the dataset and UrbanSounds[23] can improve the noise tolerance of recognition models.”
PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition (2019, methodology): “In [33] an 18-layer DaiNet with four convolutional layers in each convolutional block had achieved the best classiﬁcation performance on the UrbanSound8K dataset.”
CNN-Based Learnable Gammatone Filterbank and Equal-Loudness Normalization for Environmental Sound Classification (2020, background): “The second dataset is UrbanSound8K (US8K) [37], which is a collection of 8732 short (i.e., less than 4 seconds) excerpts of various urban sound sources.”
The Benefit of Temporally-Strong Labels in Audio Event Classification (2021, methodology): “Examples include URBAN-SED [5] which is synthesized by mixing sound events from the 10 classes of UrbanSound8K [6] using the Scaper library.”
Facilitating the Manual Annotation of Sounds When Using Large Taxonomies (2018, neutral): “More recent research on studying soundscapes shows that the taxonomical categorisation of environmental sounds is not trivial and involves many different ﬁelds, e.g., human perception or urban design [6, 7].”
SONYC (2019, background): “25,26 Our research shows that feature learning, using even simple dictionary-based methods (such as spherical k-means) makes for significant improvement in performance over the traditional approach of feature engineering.”
Addressing Missing Labels in Large-Scale Sound Event Recognition Using a Teacher-Student Framework With Loss Masking (2020, background): “As Sound Event Recognition (SER) has gained increasing attention in recent years [1], research in this ﬁeld has moved from small datasets encompassing few hours of audio [2, 3, 4, 5], to larger datasets with much greater coverage and duration [6, 7].”
Limitations of Weak Labels for Embedding and Tagging (2020, neutral): “[11] used data from UrbanSound8k [21], a synthetic dataset composed of urban sound recordings extracted from Freesound [22] and verified by human annotators.”
Chime-home: A dataset for sound source recognition in a domestic environment (2015, methodology): “Finally, we note that our annotations might further be refined to obtain a taxonomy-based annotation of source activity [15], or segmentation at the level of individual acoustic events [14].”
DCASE2017 Challenge Setup: Tasks, Datasets and Baseline System (2017, methodology): “The most used features for sound event detection are mel-scale representations, namely cepstral coefficients or log energies [12, 13, 14], and they are used with various machine learning methods, including HMMs [12], non-negative matrix factorization (NMF) [15, 16], random forests [11], and DNNs [14”
Scaper: A library for soundscape synthesis and augmentation (2017, methodology): “We used the clips from the UrbanSound8K dataset [25], approximately 1000 per each of ten urban sound sources (each clip contains one of the ten sources), as the soundbank.”
Detection and Classification of Acoustic Scenes and Events: Outcome of the DCASE 2016 Challenge (2018, background): “Simplified scenarios include having a single sound event per audio segment [13] or a sequence of non-overlapping sound events as the Office Live task in DCASE 2013 [6].”
WavCaps: A ChatGPT-Assisted Weakly-Labelled Audio Captioning Dataset for Audio-Language Multimodal Research (2023, background): “Three audio event datasets, ESC-50 [2], Ur-banSound8K [89], and VGGSound [90] ESC-50 and UrbanSound8K are ofﬁcially split into 5-folds and 10-folds for cross-validation.”
Freesound Datasets: A Platform for the Creation of Open Audio Datasets (2017, methodology): “• UrbanSound8K [30] includes a taxonomy and two audio collections for urban sounds research.”