Beschreibung: Signal-to-noise (SNR) measurement of noisy speech signals is an important topic, e. g., in automotive environments. This paper presents an approach which aims to measure the SNR of a speech signal distorted by stationary noise as close as possible to a reference-based SNR measurement according to the ITU-T Recommendation P.56, however, without using any reference signals. The proposal offers low computational complexity, and it can be applied to narrowband and wideband signals within a typical SNR range. The measured SNR is calculated by means of a separate estimate of speech and noise power, resulting in raw SNR measurements which have to be corrected by a mapping function. The proposal was evaluated thoroughly by means of speech data in 15 different languages and car noise data. The proposed method achieves a correlation coefficient larger than 0.99 compared to state-of-theart reference-based SNR measurement employing P.56 reference levels of speech and noise. The method is proposed for a new ITU-T Recommendation within Study Group 12, Focus Group Carcom.

Beschreibung: MMSE log-spectral amplitude (LSA) estimators offer convincing performance and are, therefore, widely spread in speech enhancement. In this paper an MMSE-LSA estimation under speech presence uncertainty (SPU) with a generalized gamma speech prior representing a variety of speech priors is proposed. Employing a gamma speech prior as a special case of the generalized approach, a numerical approximation is employed. Different to most state-of-the-art approaches, we choose the pdf assumptions of the mere spectral weighting rule and of the SPU estimation to be consistent. It is shown that the use of the SPU estimator with an underlying gamma speech pdf yields more than a 3 dB gain in noise attenuation, with a slight advantage versus the Gaussian SPU estimator.

Beschreibung: Signal-to-noise (SNR) measurement of noisy speech signals is an important topic in automotive environments, e.g., for investigating optimal microphone positions. This contribution presents an SNR measurement method for speech signals distorted by car noise within a typical range of SNR values. The proposed method offers low complexity and aims at providing similar results as SNR measurements based on active speech and noise levels obtained from ITU-T Recommendation P.56, however, in a reference-free fashion. "Reference-free" means that the SNR is estimated by means of only the noisy speech signal, i.e., no clean speech or noise reference signals are used. Within the measuring process, the speech and noise power are estimated separately. The SNR is then calculated as the ratio between the estimated speech and noise powers. The resulting raw SNR estimate has to be corrected by a mapping function which was trained by means of speech data in 15 different languages and car noise data. Based on speech signals with a length of 80 seconds, the proposed method achieved a maximum of 1dB absolute error in more than 90% of the test conditions and a correlation coefficient larger than 0.99 compared to state-of-the-art reference-based SNR measurement employing P.56 reference levels.