Tone emphasis device, convolutional auto encoder learning device, tone emphasis method, program

摘要

The present invention provides a tone emphasizing technology capable of emphasizing sound source with high accuracy by using DNN reflecting characteristics of musical instrument sound. SOLUTION: A frequency domain conversion unit that converts a time domain music signal into a frequency domain and generates a frequency domain music signal, and a Wiener filter estimation that estimates a Wiener filter used to emphasize a predetermined musical sound from the frequency domain music signal And a time domain conversion unit for generating an emphasized tone in the time domain from the frequency domain emphasized tone, and the Wiener filter estimation unit , Using a deep neural network including a 2 n-layer convolutional denoising auto encoder, which receives the amplitude spectrum of the music signal in the logarithmic frequency domain and outputs the amplitude spectrum of a part of the music signal in the logarithmic frequency domain And estimate the Wiener filter from the frequency domain music signal. [Selected figure] Figure 1