Freesound Datasets: A Platform for the Creation of Open Audio Datasets

By Eduardo Fonseca, Jordi Pons, Xavier Favory, Frederic Font, Dmitry Bogdanov, Andrés Ferraro, Sergio Oramas, Alastair Porter, Xavier Serra. 2017

277 citations recorded by the source corpus.

Subfields: Evaluation & Benchmarks, Speech & Audio

How researchers see this paper

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

In Evaluation & Benchmarks

In Speech & Audio

Citations in the corpus

SoundStream: An End-to-End Neural Audio Codec (2021, methodology): “For noisy speech, we synthesize samples by mixing speech from LibriTTS with noise from Freesound [51].”
Generative Speech Coding with Predictive Variance Regularization (2021, methodology): “Additional noise was added from [27], with random SNR between 0 and 40 dB SNR. Table 1 shows the combinations of coder attributes that were used.”
FSD50K: An Open Dataset of Human-Labeled Sound Events (2020, positive): “Originally released on 2017 as the Freesound Datasets platform in our previous work [56], Freesound Annotator has been object of continuous development.”
MIMII Dataset: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection (2019, background): “It is clear that the emergence of numerous open benchmark datasets [17–20] is essential for the advancement of the research field.”
Search Result Clustering in Collaborative Sound Collections (2020, methodology): “We exploit data gathered within Freesound Annotator [14] to construct 44 datasets comprising in total around 30k sounds and 215 different”
Personalized speech enhancement: new models and Comprehensive evaluation (2021, methodology): “We employed the noise ﬁles included in the Audioset [25] and Freesound [26] datasets and room impulse response (RIR) ﬁles simulated by the image method to create noisy mixtures.”
Human Listening and Live Captioning: Multi-Task Training for Speech Enhancement (2021, methodology): “The mixtures are created using 247 hours of non-stationary noise recordings from the Audioset+Freesound [26, 27] (187 hours), internal noise recordings (65 hours), and colored stationary noise (1 hour) as noise sources.”
Facilitating the Manual Annotation of Sounds When Using Large Taxonomies (2018, unknown): “In section II, we ﬁrst explain the context of this work by brieﬂy presenting the current outcome of the Freesound Datasets initiative [12], which focuses on the annotation of sounds from the Freesound database.”
Toward User-Driven Sound Recognizer Personalization with People Who Are d/Deaf or Hard of Hearing (2021, background): “Most sound recognizers both for general tasks (e.g., [11, 48]) and specifcally for DHH users [37, 38, 67] adapt deep learning approaches from computer vision—such as VGG [68] or ResNet [31]—and use transfer learning [72] to train on a large dataset of sounds such as AudioSet [26], FreeSound [25], or”
Audio tagging with noisy labels and minimal supervision (2019, methodology): “Subsequently, the candidate annotations were human-validated using a validation task deployed in Freesound Annotator , 5 an online platform for the collaborative creation of open audio datasets [23].”
Sound Event Detection in Domestic Environments with Weakly Labeled Data and Soundscape Synthesis (2019, unknown): “The foreground events are obtained from the FSD [15, 16].”
What’s all the Fuss about Free Universal Sound Separation Data? (2020, methodology): “Audio clips are of variable length ranging from 0.3 to 30s, and labels were gathered through the Freesound Annotator [21].”
Pengi: An Audio Language Model for Audio Tasks (2023, background): “[63], MELD [48]; Music Analysis: NSynth [19], FMA [9]; Audio Captioning: AudioCaps [32], ClothoV2 [15]; Audio Question and Answering: ClothoAQA [39]; Auxiliary: WavText5K [12], SoundDescs [35], MACS [42], WavCaps [43], FreeSound [20] and FindSound 2 .”
An Adversarial Feature Distillation Method for Audio Classification (2019, methodology): “Available: https://arxiv.org/abs/1807.09902 [35] E. Fonseca, J. Pons, X. Favory, F. Font, D. Bogdanov, A. Ferraro, S. Oramas, A. Porter, and X. Serra, ‘‘Freesound datasets: A platform for the creation of open audio datasets,’’ in Proc.”
Phase-aware Single-stage Speech Denoising and Dereverberation with U-Net (2020, background): “The former is a large scale dataset where the speech samples were collected from Librivox [23], and noise samples from Audioset [24] and Freesound [25].”
Disentangling speech from surroundings in a neural audio codec (2022, methodology): “We use speech waveforms from the LibriVox dataset [27], sampled at 16 kHz, and synthesize noisy speech waveforms by mixing these with noise from the Freesound dataset [28], which was screened to remove speech content and only keep clips with a CC0 license.”
Addressing Missing Labels in Large-Scale Sound Event Recognition Using a Teacher-Student Framework With Loss Masking (2020, methodology): “This process is also adopted in other large-scale datasets, e.g., [17].”
Unsupervised Contrastive Learning of Sound Event Representations (2020, methodology): “Labels in the clean and test sets are manually-labelled with Freesound Annotator [22], whereas labels in the noisy set are inferred automatically from metadata, hence featuring real-world label noise.”
Vggsound: A Large-Scale Audio-Visual Dataset (2020, methodology): “However in the audio domain, while several audio datasets have been released in the past few years [6, 7, 8, 9], the data collection process usually requires extensive human efforts, making it unscalable and often limited to narrow domains.”