In Reply To: On speech/noise correction for STI

Date Mon Jan 6 2003 22:53

Author Johan L. Nielsen (johan.nielsen@c2i.net)

> 2 Speech/noise correction: Are the speech and noise values that I put > there assumed to be those that would arise in non-reverberant > conditions? If so, why can I measure the noise level with sound level > meter? (In reverberant conditions) Can I use speech level values that I > get by subtracting the influence of the distance from the speaker to the > microphone from the standard speech spectrum values? The speech and noise levels that make up the so-called MTF correction are supposed to be measured at the receiver position, under reverberant conditions. The speech level is hence the total level; direct sound, reflections and reverberation, and no correction for distance should be made. The noise level is the ambient noise. I was really puzzled by this question at first, and had to think seriously about it. The modulation reduction index forms the basis for the STI calculation. The modulation reduction caused by reverberation is taken into account separately, in the impulse response calculation, and the two effects are combined. This is confirmed in the original papers describing the STI, for instance Houtgast, Steeneken and Plomp, "Predicting speech intelligibility in rooms from the modulation transfer function. 1. General room acoustics", Acustica vol. 46, p. 60, 1980. Another good reference describing how it can be done using MLS is Rife, "Modulation transfer function measurement with maximum-length sequences", JAES vol. 40, p. 779, 1992. It should be noted that the method of measuring separately the modulation effects of reverberation and noise will not give correct results for systems where there is signal dependent noise, like codecs. In such cases the method described by Rife using a calibrated signal, no averaging and no truncation of the response should be used, and no correction as discussed above is necessary. This is possible with the current version of WinMLS, but will be made more convenient in the next version. The method of separate measurement can be handy, for instance when the correct speech and/or noise levels cannot be obtained at the time of measurement, or when it is of interest to simulate what changes in the levels will do to the STI value. Best wishes Johan L

Subject


Your Name


Your e-mail


Message
Enter your message here.


Disk Space Available: 0K free    50,000K used


File





Click here to return to discussion list