Its too bad you can't post an image of the input signal taken from an oscilloscope as then we wouldn't need to guess. In the meantime, it would make sense to create a filter that only passes the frequency range that you expect of those pulses. So a low pass filter would be reasonable. I would bet that a lot of the noise that is getting in the way is higher frequencies than you need, so a low pass should help. The other thing that would help is to include some noise margin at the input circuit. The most common way of doing this is to include hysterisis in your slicer. With this, your input circuit would not accept that the input voltage is a high until it passes a high threshold (eg. 8 volts) and then doesn't accept that the high has gone low again until the voltage drops back down below another threshold (eg. 4 volts). Any noise that is riding on top of the input signal that is, say, 1 volt peak to peak will not trigger a transition until the input voltage gets back to within one half volt of the next threshold (for example).
Well, that's probably a poor description, but I hope you get it. One circuit that does what we need is a Schmitt Trigger. A discrete design would allow you to set your thresholds to extremes, like 2 volts and 9 volts for example, which would be alot better than using a common HC logic Schmitt trigger which has only a very small hysterisis.