| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 389.54 KB | Adobe PDF |
Advisor(s)
Abstract(s)
Esse trabalho analisa o método cepstral de cance- lamento de realimentação acústica. Demonstra-se que o erro do filtro adaptativo é composto pela soma de dois termos. O primeiro é relacionado somente ao caminho de realimentação e tende a zero, convergindo mais rapidamente para valores menores do fator de esquecimento. O segundo é relacionado aos cepstros do sinal de entrada e ao ganho do sistema de sonorização, contendo uma soma ponderada dos cepstros em que as ponderações são potências do fator de esquecimento. Simulações mostraram que, para sinais de fala, a soma ponderada dos cepstros converge, fato preponderante para a convergência do método. A velocidade de convergência, o valor após a convergência e as oscilações ao redor desse valor do segundo termo diminuem com o aumento do fator de esquecimento, comportamentos diretamente refletidos no erro do filtro. Nas primeiras iterações, o primeiro termo tem maior influência. Após algumas iterações, o segundo termo rege o desempenho do método.
This work analyzes the cepstral method for acoustic feedback cancellation. It is shown that the adaptive filter error is composed of the sum of two terms. The first is related only to the feedback path and tends to zero, converging more quickly to smaller values of the forgetting factor. The second is related to the input signal cepstra and the gain of the reinforcement system, containing a weighted sum of the cepstra where the weights are powers of the forgetting factor. Simulations showed that, for speech signals, the weighted sum of the cepstra converges, a preponderant fact for the method convergence. The convergence speed, the value after the convergence and the oscillations around this value of the second term decrease with the increase of the forgetting factor, behaviors directly reflected in the filter error. In the first iterations, the first term has the greatest influence. After a few iterations, the second term governs the method performance.
This work analyzes the cepstral method for acoustic feedback cancellation. It is shown that the adaptive filter error is composed of the sum of two terms. The first is related only to the feedback path and tends to zero, converging more quickly to smaller values of the forgetting factor. The second is related to the input signal cepstra and the gain of the reinforcement system, containing a weighted sum of the cepstra where the weights are powers of the forgetting factor. Simulations showed that, for speech signals, the weighted sum of the cepstra converges, a preponderant fact for the method convergence. The convergence speed, the value after the convergence and the oscillations around this value of the second term decrease with the increase of the forgetting factor, behaviors directly reflected in the filter error. In the first iterations, the first term has the greatest influence. After a few iterations, the second term governs the method performance.
Description
Keywords
Cancelamento de realimentação Método cepstral AFC-CE Fator de esquecimento Cepstro de fala Feedback cancellation Cepstral method Forgetting factor Speech cepstrum