Explaining deep learning models for speech enhancement - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2021

Explaining deep learning models for speech enhancement


We consider the problem of explaining the robustness of neural networks used to compute time-frequency masks for speech enhancement to mismatched noise conditions. We employ the Deep SHapley Additive exPlanations (DeepSHAP) feature attribution method to quantify the contribution of every timefrequency bin in the input noisy speech signal to every timefrequency bin in the output time-frequency mask. We define an objective metric-referred to as the speech relevance scorethat summarizes the obtained SHAP values and show that it correlates with the enhancement performance, as measured by the word error rate on the CHiME-4 real evaluation dataset. We use the speech relevance score to explain the generalization ability of three speech enhancement models trained using synthetically generated speech-shaped noise, noise from a professional sound effects library, or real CHiME-4 noise. To the best of our knowledge, this is the first study on neural network explainability in the context of speech enhancement.
Fichier principal
Vignette du fichier
dnn_explain_revised.pdf (689.39 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03257450 , version 1 (11-06-2021)



Sunit Sivasankaran, Emmanuel Vincent, Dominique Fohr. Explaining deep learning models for speech enhancement. INTERSPEECH 2021, Aug 2021, Brno, Czech Republic. ⟨10.21437/Interspeech.2021-1764⟩. ⟨hal-03257450⟩
353 View
1391 Download



Gmail Mastodon Facebook X LinkedIn More