<p>Music Melody extraction plays an important role in music creation, music education, copyright protection, and other aspects. The research focuses on the task of extracting vocal melodies from mixed music with multiple sound sources, aiming to accurately estimate the pitch sequence of the lead singer’s melody from audio containing complex sound sources such as accompaniment and harmony. However, existing Melody extraction methods face difficulties in extracting music signals from multiple sources. Therefore, an innovative melody extraction model is introduced, which leverages a graph convolutional neural network and integrates constant Q-transform with the Pearson correlation coefficient. This model first transforms music signals from the time domain to the frequency domain by exploiting the logarithmic frequency characteristics inherent in the CQT. Subsequently, it employs the Pearson correlation coefficient to quantify the interrelationships between varying frequencies and temporal intervals within the signal. Finally, by utilizing graph convolutional neural networks, the model constructs and processes signal features to effectively extract the melody. The experimental results show that the performance of the proposed model is superior to other comparative models, with extraction accuracy of over 95% on different datasets. Its pitch accuracy and frame level accuracy reach 95.85% ± 0.31% and 96.39% ± 0.28%, respectively. At the same time, this model has better separation performance for multiple sound source music signals, which is more conducive to completing melody extraction tasks and provides new melody extraction methods for multiple music related fields, promoting the development of the music industry.</p>