diff --git a/docs/index.md b/docs/index.md index 16fc39d..905794b 100644 --- a/docs/index.md +++ b/docs/index.md @@ -3,12 +3,77 @@ title: Interpreting end-to-end deep learning models for speech source localizat --- ## Manipulation of relevant input features -* ### Loc-CNN +* ### LocCNN ![Image](figures/manipulation__loc_cnn_.png) -* ### Sample-CNN +* ### SampleCNN ![Image](figures/manipulation__sample_cnn_.png) ## Feature Inspection In the following, for each environmental condition considered we will report feature results related to the best and worst performing sources for both LocCNN and SampleCNN + +### T60 0.15 s SNR 25 dB + +#### SampleCNN +* #### Best performing +![Image](figures/ex_gcc_relevance_sample_cnn_SNR_25_T60_0.15_error_min.png) + +Relevance audio signal + + +Microphone audio signal + + +* #### Worst performing +![Image](figures/ex_gcc_relevance_sample_cnn_SNR_25_T60_0.15_error_max.png) + +Relevance audio signal + + +Microphone audio signal + + +#### LocCNN +* #### Best performing +![Image](figures/ex_gcc_relevance_loc_cnn_SNR_25_T60_0.15_error_min.png) + +Relevance audio signal + + +Microphone audio signal + + +* #### Worst performing +![Image](figures/ex_gcc_relevance_loc_cnn_SNR_25_T60_0.15_error_max.png) + +Relevance audio signal + + +Microphone audio signal + +