Interpreting intermediate convolutional layers of CNNs trained on raw speech

This paper presents a technique to interpret and visualize intermediate layers in CNNs trained on raw speech data in an unsupervised manner. We argue that averaging over feature maps after ReLU activation in each convolutional layer yields interpretable time-series data. By linearly interpolating individual latent variables to marginal levels outside of the training range, we further argue that we are able to observe a causal relationship between individual latent variables that encode linguistically meaningful units and activations in intermediate convolutional layers. The proposed technique allows acoustic analysis of intermediate layers that parallels the acoustic analysis of human speech data: we can extract F0, intensity, duration, formants, and other acoustic properties from intermediate layers in order to test where and how CNNs encode various types of information. Observing the causal effect between linear interpolation and the resulting changes in intermediate layers can reveal how individual variables get transformed into spikes in activation in intermediate layers.We train and probe internal representations on two models -- a bare WaveGAN architecture and a ciwGAN extension which forces the Generator to output informative data and results in emergence of linguistically meaningful representations. Interpretation and visualization is performed for three basic acoustic properties of speech: periodic vibration (corresponding to vowels), aperiodic noise vibration (corresponding to fricatives), and silence (corresponding to stops). The proposal also allows testing of higher-level morphophonological alternations such as reduplication (copying). In short, using the proposed technique, we can analyze how linguistically meaningful units in speech get encoded in different convolutional layers.