If active speakers (i.e. speakers that require power) are connect to a computer then the amplification circuitry is a one-way street and it would be impossible for microphonic signals to be received by the computer.
If passive speakers or headphones (these are essentially the same) are connected to the computer then there is a more complicated answer.
Modern computer audio codecs often support multi-purpose audio jacks. This allows each jack to be software configured to function as: headphone output, microphone input, line out output, etc.
It is therefore possible for headphones connected to what the user presumes to be a headphone jack, to function as a microphone. Such a hack has already been demonstrated:
This hack is only possible if the headphones are not being used to output sound. It is plusable that an audio codec could be hacked to have a jack simultaneously function as a microphone input and audio output. However, any microphonic current induced will be absorbed by the output driver so that the microphonic signal is never manifested as a measurable voltage. Even if the audio output was 'silence', the output driver would still be driving a zero voltage signal and absorbing microphonic current.
In summary: Active speakers cannot be used for listening/surveillance. Passive speakers or headphones can be used for listening/surveillance but only if they are not being used to output audio.