Modeling speech and audio codecs reverberation artifact

Speech and sound codecs subjective assessment requires anchor signals to allow the comparison of results from different laboratories. The anchor signal presently used, the Modulated Noise Reference Unit (MNRU), is based on the hypothesis that coding technique artifact is only due to quantization noise. Earlier work showed that the impairments of speech codecs could be described by a four-dimensional event and two of these dimensions, namely “Muffled” and “Background noise” dimensions, have been already modeled. In this paper, we propose to design anchor signals for the third dimension mostly characterized by “Echo/Reverberation” attribute.