oh you could do it in software only, but its going to be a royal ....
here is the basic idea
the arduino can really only have 2 states on a pin, on or off, a speaker is a analog device it can be anywhere tween fully out and fully at rest SO, what you would need to do is get your sounds, 8 bit, 11khz at best, chop off the wav file header and store it in rom (see progmem in the reference) these are going to need to be extremely short due to the limited space
next your going to need a carrier, something that keeps the speaker about in mid flux so plan on generating ~22khz or so, or else your going to hear a high pitched squeak, but not so fast or slow that the sound is going to be played back too fast or slow .. or at all
then your going to need to write up a software DAC, 4-8 bit to convert the data from binary form to an analog value and have that output as PWM on your carrier all while not using every single ounce of power in the chip
not impossible but not exactly a "going to whip it out in 5 hours" type of deal, especially if you have never reproduced audio on this low of a level on this limited hardware
good luck
and no if you think your going to use hardware there is nothing stopping you from going ahead and writing software, though you wont fully know if its working until everything is in place