Estimating Number of Speakers via Density-Based Clustering and Classification Decision

Abstract

It is crucial to robustly estimate the number of speakers (NoS) from the recorded audio mixtures in a reverberant environment. Some popular time-frequency (TF) methods approach this NoS estimation problem by assuming that only one of the speech components is active at each TF slot. However, this condition is violated in many scenarios where the speeches are convolved with long length of room impulse response coefficients, which causes degenerated performance of NoS estimation. To tackle this problem, a density-based clustering strategy is proposed to estimate NoS based on a local dominance assumption of speeches. Our method consists of several steps from clustering to classification of speakers with the consideration of robustness. First, the leading eigenvectors are extracted from the local covariance matrices of mixture TF components and ranked by the combination of local density and minimum distance to other leading eigenvectors with higher density. Second, a gap-based method is employed to determine the cluster centers from the ranked leading eigenvectors at each frequency bin. Third, a criterion based on averaged volume of cluster centers is proposed to select reliable clustering results at some frequency bins for the classification decision of NoS. The experiment results demonstrate that the proposed algorithm is superior to the existing methods in various reverberation cases with noise-free condition or noise condition. INDEX TERMSNumber of speakers, speeches, reverberation, audio source separation (ASS), local dominance, density-based clustering. YI GUO (M'16) received the B.E. degree (Hons.

Estimating number of group targets arranged in linear array via forward‐backward matrix pencil method

“…In a real scenario, the radar echoes are mixed signals that contain amplitude and phase information for each target. The traditional mixed signal obtained based on a mixing matrix only contains amplitude information, but not phase information [16]. Therefore, the processed radar echo in this paper is closer to the actual mixed signal compared with the classical parameter estimation method. From the view of solving process, the computation complexity of this proposed method is similar compared with the traditional direction of arrival estimation method (Such as Akaike information criterion (AIC), minimum description length (MDL)) [27].…”

“…In a real scenario, the radar echoes are mixed signals that contain amplitude and phase information for each target. The traditional mixed signal obtained based on a mixing matrix only contains amplitude information, but not phase information [16]. Therefore, the processed radar echo in this paper is closer to the actual mixed signal compared with the classical parameter estimation method.…”

“…[15] proposed a novel strategy inspired by the jackknifing resampling method to improve source number detection and direction estimation. Besides that, a density‐based clustering strategy was proposed to estimate the number of speakers based on a local dominance assumption of speeches [16]. In addition, for the sparse signals in the SCBSS, the matrix pencil method (MPM) also can be used to estimate the two‐dimensional parameters effectively [17, 18].…”

An ASC‐Based Unitary Matrix Pencil Method for Parameter Estimation of Group Target

“…Because the echo information in this paper contains amplitude and phase information. The signal separation technology is based on the mixing matrix to get the target echo (Yang et al., 2019), which does not contain phase information but only amplitude information. The proposed method in this paper has a wider application range compared with other matrix pencil methods, such as forward‐backward matrix pencil method (FBMPM). FBMPM is only applicable to the targets arranged in linear array (Sun et al., 2022).…”

“…Because the echo information in this paper contains amplitude and phase information. The signal separation technology is based on the mixing matrix to get the target echo (Yang et al, 2019), which does not contain phase information but only amplitude information. 2.…”