Welcome to A Canine Language Lexical Analysis System!
Enjoy the system demonstration video below:
On the 'Phonemes' page, we have projected the clustering centers of 50 different phonemes onto a two-dimensional diagram. You can click on different phoneme points to listen to samples and perceive the distinctions between various phonemes.
On the 'Vocabulary' page, we have provided a subset of the most plausible vocabulary words list. You can click on different words to access audio samples of that word and the corresponding video clips where the sound sample is located. In order to provide a clearer context of a word, we have extended the length of the video both before and after the sample 0.5 second.
On the 'Test your data' page, you can upload your own recorded Dog Voice, which should be less than three seconds in length. Afterward, select the model you wish to run from the dropdown menu.
'Original' refers to the energy graph of the audio itself, and you can play the complete audio.
'Acoustic' describes the audio segmentation method based on energy trends in the audio, as detailed in a 2023 paper. This method assigns labels to the segments based on the International Phonetic Alphabet (IPA) that most closely matches the mel features of the audio. The buttons labeled with IPA symbols below can be clicked to listen to the segmented audio pieces.
'Hubert' is the segmentation method proposed in our paper, where the labels do not have specific meanings; identical labels indicate the same phonemes. The following graph displays a scatter plot of the centroids for each label class, reduced to two dimensions using Principal Component Analysis (PCA). It roughly indicates that audio files marked with the same label exhibit similar acoustic expressions.
On the 'examples' page, you can select from 30 randomly chosen dog voice audio clips from the test dataset via the dropdown menu and view the segmentation results using both the audio-based method and our proposed method. Similarly, you can click the buttons labeled with labels below to listen to the segmented audio pieces.