Sound Event Detection and Separation: a Benchmark on Desed Synthetic Soundscapes

Nicolas Turpault (1) , Romain Serizel (1) , Scott Wisdom (2) , Hakan Erdogan (2) , John R Hershey (2) , Eduardo Fonseca (3) , Prem Seetharaman (4) , Justin Salamon (5)

1 MULTISPEECH - Speech Modeling for Facilitating Oral-Based Communication
2 Google Inc
3 UPF - Universitat Pompeu Fabra [Barcelona]
4 Descript, Inc.
5 Adobe Research

We propose a benchmark of state-of-the-art sound event detection systems (SED). We designed synthetic evaluation sets to focus on specific sound event detection challenges. We analyze the performance of the submissions to DCASE 2021 task 4 depending on time related modifications (time position of an event and length of clips) and we study the impact of non-target sound events and reverberation. We show that the localization in time of sound events is still a problem for SED systems. We also show that reverberation and non-target sound events are severely degrading the performance of the SED systems. In the latter case, sound separation seems like a promising solution.

Mots clés: Sound event detection; synthetic soundscapes; sound separation

Domaines: Son [cs.SD]; Traitement du signal et de l'image [eess.SP]

Nicolas Turpault, Romain Serizel, Scott Wisdom, Hakan Erdogan, John R Hershey, et al.. Sound Event Detection and Separation: a Benchmark on Desed Synthetic Soundscapes. ICASSP 2021 - 46th International Conference on Acoustics, Speech, and Signal Processing, Jun 2021, Toronto/Virtual, Canada. ⟨10.1109/ICASSP39728.2021.9414789⟩. ⟨hal-02984675⟩