For starters, I don't recommend using Sound Forge to do what you are attempting. Compared to other DAWs, it is limited.
What I would do if possible is to have the volume change be gradual instead of instantaneous. In every other DAW this is done by applying an automation or envelope to the gain/volume with either a curve or a simple straight line which represents how much it changes over the allotted time frame. Looks like this in Logic, like this in Ableton, and like this in FL Studio. This may require you to zoom in and make this gradual transition by hand, which would be a huge pain in the butt.
The closet thing I found after a googling was adjusting the volume envelope in Sound Forge:
Also take note, having such a vast difference in volume (90% change!) is making the smoothness of the transition much more noticeable.
If you want to completely separate the speech from the audio without any of the song in the background, there are a few different options as well. The first is through a technique which requires you to have an exact copy of the song but in instrumental form. This technique is explained in this eHow tutorial. While doable in Sound Forge with this technique, I think this is easier to do in Audacity (which is free) though. The quick how to of that is to invert the wave signature of the instrumental track and overlay it onto the original making the audio signals cancel each other out leaving only the vocal elements.
My favorite way of doing this is using the amazing software known as Melodyne because of its ease of use that the fact that it was designed to do tasks like this. Melodyne is freaking awesome at isolating different parts of songs and audio files.
IF I HAD TO DO IT, I WOULD DO IT WITH MELODYNE BECAUSE IT WOULD COMPLETELY REMOVE THE BACKGROUND SONG.
Lastly, if you are extremely limited in your options, you could always do it with Audacity since its free. Here is a video explaining how to use the volume envelope.
If you are completely set on using Sound Forge here are some videos explaining how to do what you ask: video 1, video 2.