Well, sip dtmf is usually detected and sent out of band, in call dtmf over TDM/analog is of course constrained to in-band. With reference to
you could simplistcally run your recording through a single pass three element filter that bandrejects "1209 Hz" "1336 Hz" and "1477 Hz" this should effectively suppress any spuriously detected DTMF signaling, the width and height of the filter would have to be experimented with, using sox you can "trial and error" with
play yourfile rate 16000 sinc -t 8 -a 180 1214-1204 sinc -t 8 -a 180 1341-1331 sinc -t 8 -a 180 1482-1472
where -t is the total band width and -a the attenuation, the range should symetrically span the single sine frequency of interest. Reverse the range for a bandpass to hear them one by one, you should "man sox" first so you fully understand.
If you are comfortable with deploying python code then probably:-
will likely be a lot more effective and less "disturbing" as it uses the slower but more precise DFT (discrete fourier transform) extractions, (sorry, possibly too much math theory here 
but if it worked for AT&T . . .)
A caveate is that I can't say whether that will fix your problem and ANY filter will "disturb" the audio.