Narrow Results

In order to improve the accuracy of music post-production annotation and the score of music post-production automatic annotation, this paper proposes a music post-production automatic annotation method based on a hybrid Boltzmann machine (HBM) and pitch features. First, obtain the post-production music label dataset, and obtain the song theme label dataset and song sentiment label dataset through statistical analysis. Second, the dataset is partitioned using a dataset partitioning method and stored in MP3 format. Then, the discrete Fourier transform (DFT) is used to process the frames of the audio to obtain the spectrum, and a mixed Boltzmann machine is used to perform low-pass filtering and downsampling on each subband to obtain the amplitude envelope of the audio on the time axis. The first five peaks obtained are added together to obtain a beat histogram, which enables the extraction of pitch features for later production music. Finally, the music sequence is modeled using the BRNN network structure, and the attention weights of post-production music sequence features are calculated. The Mel score of the song is preprocessed, and the prediction probability of the song’s music labels is obtained through music classification methods. The configuration information of the labels is read. The labels are filtered based on the threshold in the label configuration information, the final labels are stored in the database, and an automatic annotation method is implemented for music post-production. The results show that the annotation accuracy of our method can reach 99.97%, and the annotation score can reach 0.98. Our method has stronger robustness.

Biomimetic robots borrow their senses and structure from animals, such as insects, fishes, and human. Development of underwater vehicles is one of the areas where biomimetic robots can potentially perform better than conventional robots. In this paper, an undulating fin mechanism has been developed and used as the propulsion system of fish in various fin types. The layout and workspace of the modular fin segments are considered and analyzed. The relationship of the individual fin segment and phase angles with the overall fin trajectory is also discussed. A gymnotiform knifefish robot, as an example, has been developed to demonstrate the design methodology and prototype performance. The maneuvering and the buoyancy control can be achieved by the integration of a buoyancy tank with the undulating fin. Experiments were conducted in the laboratory tank and the variation of velocity with respect to several swimming parameters was analyzed. Field trials have also been conducted in an outdoor pool to demonstrate the swimming capability of the knifefish robot and its buoyancy performance in 4 m deep water.