I would like to respond to Maurice and previous comments on double talk detection. I worked for several years in speakerphone design and development. There are MANY situations where the double talk signal can be significantly hotter than the local talker. As Maurice suggests, the algorithms that have been used for double talk detection in network echo cancellation just do not apply to acoustic echo cancellers. A brief discussion to justify this and the idea of echoes being hotter than the local talker may be in order. The ITU defined a standard P.34 (if my memory serves me right) that defines the level requirements for a speakerphone attached to the telephone network. I realise that this is not where some of you are working, but it is the intended environment for many designs being worked. It defines parameters for the transmit signal and receive signal on terms called Objective Loudness Rating, commonly known as TOLR and ROLR. The standard specifies a typical environment in which a speakerphone would be used. It assumes the speakerphone is on a desk 40 cm from the front edge, and that the user is immediately above the front edge of the desk, at 30 cm above. We now have a 30, 40, 50 cm triangle. So the user is 50 cm from the instrument. In principle, the level heard by a user with nominal volume at the speakerphone, should be the same as if he/she had a handset pressed against the ear. Also the gain of the microphone should be such that the level on the telephone line is the same as if the user were using a handset (say 2 cm from the mouth). Actually, there is some account taken for the fact that the average user tends to shout at a speakerphone. Now consider that the speaker and microphone in the unit may be as close as 2-3 cm from each other. That means that they are literally shouting at each other and there is no ear seal to minimize the "coupling" between them. It is clear that there can easily be considerable positive gain between the speaker and microphone withthem being 3cm apart and the user being 50 cm away from both of them. Now consider that two users may have the same kind of instruments. There is now a loop that includes two sections with considerable positive gain (I have seen in excess of +12dB) at each end. Double talk detection and residual echo removal, after the canceller, are issues that need great care and attention. I hope these few comments may have been useful to someone out there. Regards, Dave Shaw To answer specifically your question, the echo contains the double-talk signal as well as the echo. By applying the ALC to this signal, you are amplyfing the echo as well as the double-talk signal. The ALC my not help since the ratio of the 2 will not be changed. In the worst case, the echo can be 6 dB down from a nominal input level, and the double-talk signal can be 8 dB down from a nomimal input level. Under these conditions, you need more than level detection to determine if double-talk is present. Maurice Givens |