Re: AGC on a phone conversation
Andras Kadinger <[email protected]>
| Newsgroups | gmane.comp.audio.compression.speex.devel |
|---|---|
| Message-ID | <[email protected]> |
2011.08.23. 15:38 keltezéssel, Yanick Bourbeau írta: > Since I don't have access to different channels as I record a phone call > using a man in the middle approach, there is something else I can use > to equalize the sound ? What I would do then probably is try to manually separate the two channels/talkers; say channel 1 goes from 0 seconds to 13 seconds, channel 2 goes from 13 seconds to 27 seconds, then channel 1 again goes from 27 seconds to ... you understand. Then you have two channels, one channel is always silent/mute, the other contains the current talker. Then you can AGC them separately. As described here, this is a manual process. So if you have a lot of audio you need to process, manual might not make sense anymore, and you could then try to create an algorithm to do this automatically. I would make some statistics about the loudness (signal level) of the sides, and the pauses they make, and try to come up with an algorithm that analyzes the whole call, and tries to segment it into the two sides, based on the high probability that signal level usually stays the same or only changes slowly within the same channel, but it is considerably different between the two channels/talkers. Alternatively, you could just try to adjust the parameters of the AGC to provide for a faster adjustment so that it catches up relatively quickly whenever the participants change roles (one stops talking and the other one starts talking - much louder/quieter) - you might be able to find a set of AGC settings that work acceptably for your calls.