In this application-oriented paper we consider the problem of elimination of impulsive disturbances, such as clicks, pops and record scratches, from archive audio recordings. The proposed approach is based on bidirectional processing-noise pulses are localized by combining the results of forward-time and backward-time signal analysis. Based on the results of specially designed empirical tests (rather than on the results of theoretical analysis), incorporating real audio files corrupted by real impulsive disturbances, we work out a set of local, case-dependent fusion rules that can be used to combine forward and backward detection alarms. This allows us to localize noise pulses more accurately and more reliably, yielding noticeable performance improvements, compared to the traditional methods, based on unidirectional processing. The proposed approach is carefully validated using both artificially corrupted audio files and real archive gramophone recordings.

Elimination of Impulsive Disturbances From Archive Audio Signals Using Bidirectional Processing

Gramophone records were the main recording medium for more than seven decades and regained widespread popularity over the past several years. Being an analog storage medium, gramophone records are subject to distortions caused by scratches, dust particles, degradation, and other means of improper handling. The observed noise often leads to an unpleasant listening experience and requires a filtering process to remove the unwanted disruptions and improve the audio quality. This paper proposes a novel approach that employs various feed forward time delay artificial neural networks to detect and reconstruct noise in musical sound waves. A set of 800 songs from eight different genres were used to validate the performance of the neural networks. The performance was analyzed according to the outlier detection and interpolation accuracy, the computational time and the tradeoff between the accuracy and the time. The empirical results of both detection and reconstruction neural networks were compared to a number of other algorithms, including various statistical measurements, duplication approaches, trigonometric processes, polynomials, and time series models. It was found that the neural networks’ outlier detection accuracy was slightly lower than some of the other noise identification algorithms, but achieved a more efficient tradeoff by detecting most of the noise in real time. The reconstruction process favored neural networks with an increase in the interpolation accuracy compared to other widely used time series models. It was also found that certain genres such as classical, country, and jazz music were interpolated more accurately. Volatile signals, such as electronic, metal, and pop music were more challenging to reconstruct and were substantially better interpolated using neural networks than the other examined algorithms.

Gramophone Noise Detection and Reconstruction Using Time Delay Artificial Neural Networks

In this paper the problem of detection of impulsive disturbances in archive audio signals is considered. It is shown that semi-causal/noncausal solutions based on joint evaluation of signal prediction errors and leave-one-out signal interpolation errors, allow one to noticeably improve detection results compared to the prediction-only based solutions. The proposed approaches are evaluated on a set of clean audio signals contaminated with real click waveforms extracted from silent parts of old gramophone recordings.

Detection of impulsive disturbances in archive audio signals

Gramophones were the main audio recording medium for more than seven decades and regained widespread popularity over the past few years. Being an analogue storage medium, gramophone records are subject to distortions caused by scratches, dust particles, high temperatures, excessive playback and other noise induced by mishandling the record. Due to the early adoption of the compact disc and other digital audio mediums, most research to reduce the noise on gramophone records focused on physical improvements such as the enhancements of turntable hardware, amelioration of the record material or advances through better record cutting techniques. Comparatively little research has been conducted to digitally filter and reconstruct distorted gramophone recordings. This thesis provides an extensive analysis on the digital detection and reconstruction of noise in gramophone audio signals distorted by scratches. The ability to approximated audio signals was examined though an empirical analysis of different polynomials and time series models. The investigated polynomials include the standard, Fourier, Newton, Lagrange, Hermite, osculating and piecewise polynomials. Experiments were also conducted by applying autoregressive, moving average and heteroskedasticity models, such as the AR, MA, ARMA, ARIMA, ARCH and GARCH models. In addition, different variants of an artificial neural network were tested and compared to the time series models. Noise detection was performed using algorithms based on the standard score, median absolute deviation, Mahalanobis distance, nearest neighbour, mean absolute spectral deviation and the absolute predictive deviation method. The reconstruction process © University of Pretoria employed the examined polynomials and models and also considered adjacent window, mirroring window, nearest neighbour, similarity, Lanczos and cosine interpolation. The detection and reconstruction algorithms were benchmarked using a dataset of 800 songs from eight different genres. Simulations were conducted using artificially generated and real gramophone noise. The algorithms were compared according to their detection and reconstruction accuracy, the computational time needed and the tradeoff between the accuracy and time. Empirical analysis showed that the highest noise detection accuracy was achieved with the absolute predictive deviation using an ARIMA model. The predictive outlier detector employing a Jordan simple recurrent artificial neural network was most efficient by achieving the best detection rate in a limited timespan. It was also found that the artificial neural networks reconstructed the audio signals more accurately than the other interpolation algorithms. The AR model was most efficient by achieving the best tradeoff between the execution time and the interpolation error.

Digital Audio Restoration of Gramophone Records

Autoregressive models are widely employed for predictions and other inferences in many scientific fields. While the determination of their order is in general a difficult and critical step, this task becomes more complicated and crucial when the time series under investigation is realization of a stochastic process characterized by sparsity. In this paper we present a method for order determination of a stationary AR model with a sparse structure, given a set of observations, based upon a bootstrapped version of MAICE procedure [Akaike H. Prediction and entropy. Springer; 1998], in conjunction with a LASSO-type constraining procedure for lag suppression of insignificant lags. Empirical results will be obtained via Monte Carlo simulations. The quality of our method is assessed by comparison with the commonly adopted cross-validation approach and the non bootstrap counterpart of the presented procedure.

LASSO order selection for sparse autoregression: a bootstrap approach

This paper presents a new approach to elimination of impulsive disturbances from archive speech signals. The proposed sparse autoregressive (SAR) signal representation is given in a factorized form — the model is a cascade of the so-called formant filter and pitch filter. Such a technique has been widely used in code-excited linear prediction (CELP) systems, as it guarantees model stability. After detection of noise pulses using linear prediction, the factorized model is converted into a generic sparse form in order to perform a projection-based signal interpolation. It is shown that the proposed algorithm is able to deal favorably with speech signals with strong glottal activity, which is a serious problem for algorithms based on the classical AR modeling.

/pdf/elimination-of-clicks-from-archive-speech-signals-using-1312g420r5.pdf

Marcin Ciotek

Papers

Elimination of clicks from archive speech signals using sparse autoregressive modeling