Write yourself a script which, when initiated, simultaneously triggers the EEG recording software to start recording and some sound recording software to start recording audio. Place a microphone close to some speakers/headphones and have the experiment run it's course. Once finished, one has to assume that the audio recording and the EEG data are time synced (make sure to consider sampling rates and what not). Open your audio file in matlab and use some peak detection algorithm (findpeaks()) to find the onsets of each of your stimuli.

This might be difficult if there are subtle differences between stimuli across different conditions, and you need to identify which condition a trial belongs to. If this is the case, then, if you can, have your experiment present a uniquely shaped tone for each condition through some headphones. Tape a microphone on to the headphones, and then load this audio file into matlab. Here, you can extract the onset times of these tones and observe their shape within the timeseries, thus allowing you to identify which condition the presented stimulus belonged to.

Easiest and cheapest? Just use one channel of EEG as a sound recorder and analyse all together on common waveforms

You can use the Matlab Psychtoolbox and/or Data Acquisition Toolbox to record from your audio port:

http://psychtoolbox.org/docs/PsychPortAudio

https://nl.mathworks.com/matlabcentral/answers/24502-speech-signal-from-microphone-usb