Informed audio source separation: A comparative study

International Journal of Adaptive Control and Signal Processing, 2004

The problem of separating out a number of audio sources observed from an array of microphones in a real room environment has received a great deal of attention in the past decade. While there are now a number of workable methods that can even deal with relatively high reverberation [18], a number of interesting problems still remain. In this paper, the authors review the methods based around Independent Component Analysis (ICA), discussing the various choices available in algorithm design. We then explore the issue of sensitivity to speaker movement which appears to impose fundamental limitations on BSS performance.

2010

This paper introduces the audio part of the 2010 community-based Signal Separation Evaluation Campaign (SiSEC2010). Seven speech and music datasets were contributed, which include datasets recorded in noisy or dynamic environments, in addition to the SiSEC2008 datasets. The source separation problems were split into five tasks, and the results for each task were evaluated using different objective performance criteria. We provide an overview of the audio datasets, tasks and criteria. We also report the results achieved with the submitted systems, and discuss organization strategies for future campaigns.

Wseas Transactions on Signal Processing, 2008

Blind source separation is an advanced statistical tool that has found widespread use in many signal processing applications. However, the crux topic based on one channel audio source separation has not fully developed to enable its way to laboratory implementation. The main idea approach to single channel blind source separation is based on exploiting the inherent time structure of sources known as basis filters in time domain that encode the sources in a statistically efficient manner. This paper proposes a technique for separating single channel recording of audio mixture using a hybrid of maximum likelihood and maximum a posteriori estimators. In addition, the algorithm proposes a new approach that accounts for the time structure of the speech signals by encoding them into a set of basis filters that are characteristically the most significant.

This paper proposes a new informed source separation technique which combines music transcription with source separation. The presented system is based on a coder/decoder configuration where a classic (not informed) multiple-F0 estimation is applied on each separated source signal assumed known at the coder before the mixing process. Thus, the extra information required to recover the reference transcription of each isolated instrument is computed and inaudibly embedded into the mixture using a watermarking technique. At the decoder, where the original source signals are unknown, instruments are separated from the mixture using the informed transcription of each source signal. In this paper, we show that a classic (non-informed) F0 estimator can be used to reduce the amount of bits necessary to transmit the exact transcription of each isolated instrument.

Speech and Audio Processing, IEEE …, 2005

In this paper we propose a new frequency domain approach to blind source separation (BSS) of audio signals mixed in a reverberant environment. It is first shown that joint diagonalization of the cross power spectral density matrices of the signals at the output of the mixing system is sufficient to identify the mixing system at each frequency bin up to a scale and permutation ambiguity. The frequency domain joint diagonalization is performed using a new and quickly converging algorithm which uses an alternating least-squares (ALS) optimization method. The inverse of the mixing system, estimated using the joint diagonalization algorithm, is then used to separate the sources. An efficient diadic algorithm to resolve the frequency dependent permutation ambiguities that exploits the inherent non-stationarity of the sources is presented. The effect of the unknown scaling ambiguities is partially resolved using a novel initialization procedure for the ALS algorithm.

Demixing consists in recovering the sounds that compose a multichannel mix. Important applications include karaoke or respatialization. Several approaches to this problem have been proposed in a coding/decoding framework, which are denoted either as spatial audio object coding or informed source separation. They assume that the constituent sounds are available at an encoding stage and used to compute a side-information transmitted to the end-user. At a decoding stage, only the mixtures and the side information are used to recover the sources. Here, we propose an advanced model, which encompasses many practical scenarios and permits to reach bitrates as low as 0.5kbps/source. First, the sources may be mono or multichannel. Second, the mixing process is not assumed to be linearinstantaneous or convolutive as is usual, but rather diffuse, permitting professional mixes to be processed. Third, the signals to be recovered may either be the original sources or their images.

IEEE International Symposium on Signal Processing and Information Technology, 2009

Several algorithms for instantaneous blind source separation (BSS) have been introduced in the past years. The performance of these algorithms needs to be evaluated and assessed to study their merits and choose the best of them for a given application. In this paper, a new adaptive approach is presented to evaluate different blind source separation algorithms. In this new approach, three new evaluation metrics are added. The first metric is the minimum number of samples required for a successful separation process. The second metric is the time needed to complete the separation process. The third metric is the number of sources that the BSS algorithm can separate from their mixtures. The new approach is used to compare three different blind source separation algorithms. These algorithms are: kurtosis, negentropy, and the maximum likelihood. Since the evaluation of a BSS technique is application-dependent, we are using the same application (separation of audio sources) to evaluate each of these BSS algorithms. The comparison, between the three algorithms, shows that the maximum likelihood has the best performance and the kurtosis is the faster. This motivates us to develop a new hybrid approach that combines the two algorithms to gain the benefits from both algorithms. In this new algorithm we start with the maximum likelihood (ML) algorithm to find the separation matrix and then tune this matrix by the kurtosis algorithm.

IEEE Transactions on Speech and Audio Processing, 2003

The problem of separation of audio sources recorded in a real world situation is well established in modern literature. A method to solve this problem is Blind Source Separation (BSS) using Independent Component Analysis (ICA). The recording environment is usually modelled as convolutive. Previous research on ICA of instantaneous mixtures provided solid background for the separation of convolved mixtures. The authors revise current approaches on the subject and propose a fast frequency domain ICA framework, providing a solution for the apparent permutation problem encountered in these methods.

2012 Proceedings of the 20th European Signal Processing Conference (EUSIPCO), 2012

Recent advances in multimedia technology opened the path for individual manipulation of the different audio objects within a multichannel mix, for both sampling and karaoke applications. This requires the transmission of these objects as an additional information. Informed Source Separation (ISS) is an adequate framework for this problem. Its main idea is not to transmit the objects themselves, but rather the parameters required to recover them using the mixtures and separation algorithms. In recent studies, the connection was made between ISS and source coding and the concept of coding-based ISS (CISS) was introduced. CISS differs from classical source coding in its use of the mixtures, which permits to reduce the bitrates required to convey audio objects compared to source coding alone with the same model. In this study, we extend existing work on CISS to the case of multichannel mixtures and demonstrate a considerable increase of performance over classical ISS.

IEEE Transactions on Audio, Speech, and Language Processing, 2000

This paper presents a technique for Informed Source Separation (ISS) of a single channel mixture, based on the Multiple Input Spectrogram Inversion method. The reconstruction of the source signals is iterative, alternating between a timefrequency consistency enforcement and a re-mixing constraint. A dual resolution technique is also proposed, for sharper transients reconstruction. The two algorithms are compared to a state-of-the-art Wiener-based ISS technique, on a database of fourteen monophonic mixtures, with standard source separation objective measures. Experimental results show that the proposed algorithms outperform both this reference technique and the oracle Wiener filter by up to 3dB in distortion, at the cost of a significantly heavier computation.