Research on Stage Vocal Classification Based on Long Short-Term Memory Artificial Neural Networks

With the continuous advancement of artificial intelligence technology, stage vocal classification has become increasingly prevalent in the fields of music and speech processing. In response to the complex sound data and realtime classification requirements in stage performance environments, this paper proposes a stage vocal classification method based on Long Short-Term Memory (LSTM) neural networks. First, the paper deeply analyzes the audio features of stage vocals, focusing on extracting parameters closely related to vocal characteristics, such as Mel-Frequency Cepstral Coefficients (MFCC), formant frequencies, pitch, and timefrequency domain features. These features can comprehensively capture both temporal and spectral information of the human voice, providing effective input for the LSTM model. Secondly, considering the advantages of LSTM networks in handling time-series data, the paper constructs an LSTM-based stage vocal classification model that leverages its ability to learn temporal dependencies and sequence patterns. This model not only accurately captures pitch variations, speech rhythm, and environmental noise interference in stage performances but also effectively distinguishes between different singing styles. Experimental results show that the proposed method outperforms traditional machine learning methods in stage vocal classification accuracy, especially in complex stage environments, achieving higher classification accuracy and robustness.