核心概念
The author evaluates autoregressive audio inpainting methods, highlighting the importance of AR model estimators and model order in achieving high-quality results.
要約
The content discusses the evaluation of popular audio inpainting methods based on autoregressive modeling. It compares extrapolation-based and Janssen methods, introducing a novel variant of the Janssen method for gap inpainting. The paper emphasizes the significance of AR model estimators and model orders in achieving optimal results. The study includes experiments on an audio inpainting dataset to showcase the performance of different approaches. Key metrics like SDR and ODG are used to assess the quality of reconstructed audio signals. The computational aspects, including speed and algorithm complexity, are also discussed in detail.
統計
"For a human listener, the result should be as pleasant as possible and ideally not noticeable."
"The results demonstrate the importance of the choice of the AR model estimator and the suitability of the new gap-wise Janssen method."
"The perceived quality of the signal is evaluated using the objective metric PEMO-Q."
"The quality of reconstructed audio is assessed using the signal-to-distortion ratio (SDR)."
"We chose solo instruments since AR models are expected to perform well on them."
"To simulate degradation, we consider gap lengths from 10 ms up to 80 ms."
"The computational load is proportional both to the order of the AR model and to the gap length."
"For all approaches, computational load is proportional both to order of AR model and gap length."
"Burg algorithm is more demanding compared to LPC."
"Elapsed times are up to around 0.15 s per signal with p = 2048, while gap-wise Janssen reaches up to 11.5 s per signal with p = 1024."
引用
"The main differences between particular popular approaches are pointed out, and a mid-scale computational experiment is presented."
"The experiments demonstrated the importance of choosing between LPC or Burg algorithm for AR model estimation."
"The concluding test revealed that gap-wise Janssen method using Burg algorithm is recommended as an autoregressive reference for future tests on inpainting middle-length gaps."