Separación de la voz cantada en señales musicales monocanal, utilizando técnicas NMF aplicadas a espectrogramas de diferente resolución.
Fecha
2016-06-07
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
[ES] Este Trabajo Fin de Grado está centrado en el procesado de audio, concretamente en la separación de fuentes sonoras. El alumno deberá implementar un sistema que sea capaz de separar la señal vocal (singing-voice) procedente del cantante del resto del acompañamiento musical (instrumentos musicales). Para tal fin, se utilizarán técnicas de descomposición de matrices no negativas (NMF) aplicadas al espectrograma de la señal musical con diferente resolución tiempo-frecuencia. Además, se implementará un método para discernir continuidad espectral y temporal lo cual permitirá descomponer la señal instrumental en armónica y percusiva. El proyecto se implementará utilizando el entorno MATLAB.
[EN] This thesis is focused on audio processing, namely in the sound sources separation. The student must implement a system which will be able to separate the speech signal (singing voice) originating from the singer, from the rest of the musical accompaniment (musical instruments). To this end, non-negative matrix factorization techniques (NMF) will be used, applying them to the music signal spectrogram with different time-frequency resolution. In addition, a method to discern spectral and temporal continuity will be implemented, which will allow to separate the instrumental signal into pitched and percussive signal. The project will be implemented using the MATLAB environment.
[EN] This thesis is focused on audio processing, namely in the sound sources separation. The student must implement a system which will be able to separate the speech signal (singing voice) originating from the singer, from the rest of the musical accompaniment (musical instruments). To this end, non-negative matrix factorization techniques (NMF) will be used, applying them to the music signal spectrogram with different time-frequency resolution. In addition, a method to discern spectral and temporal continuity will be implemented, which will allow to separate the instrumental signal into pitched and percussive signal. The project will be implemented using the MATLAB environment.
Descripción
Palabras clave
Sistemas de Telecomunicación/Sonido e Imagen