I recently recorded some subjects reading lists of words and sentences. The recordings were taken on a ZOOM H4N recorder, sampled at 44100HZ, on a head-mounted condenser microphone.
The recordings would be high-quality if there were no echo. I want to do forced-alignement on these recordings and measure intensity as a correlate of the lenition of voiced stops.
How bad are echoes for taking measurements? Is there a way to filter the echo out?
Well, during the closure of a voiceless stop, the intensity should be zero, but with an echo, it will be that plus whatever the speaker said some time(s) prior, so in "pot", the amplitude of "t" will include a trace of the vowel. I would be surprised if filtering didn't mutilate the amplitude of the signal to the point that your results would be questionable. If you are simply hoping to show e.g. that amplitude is lower when a voiced stop is lenited, you have independent control over whether a stop is lenited, and can be certain that in the context where there is lenition there is not coincidentally a co-varying difference in amplitude on the preceding segment (within 500 msc), then you may still be able to make the case for a categorial correlation, though the actual amplitude values would not be any use.
