I have an old video. It was recorded in 1995 and somebody is reading a letter loud while music in de background was playing. It is hard to understand what the person is saying. Can this be improved?
A difficult task considering that the 1995 audio may have been recorded on VHS/BETA if a domestic recording - and not much better if on U-Matic or the like.
You could try using EQ in SoundForge Pro to reduce the levels of the music frequencies outside of the voice frequency range, but don't expect too much especially if the music frequency range is in that same range.
iZotope's RX8 Advanced has a Dialogue Isolate function, but RX8 Adv is very expensive. Steinberg's SpectraLayers Pro 7 (less expensive) has a brilliant function that allows sections of the audio to be broken down into stems, including dialogue. But depending on the quality of the recording and the difference in dialogue v background/music levels, the result may be only marginal and if pushed too far will likely result in audio artifacts that make the process worse than the original. The reason is that the dialogue frequencies are likely to be in the same frequency range as is some of the music's frequencies - unless the music frequencies are in the high range like violins and don't compete with the voice's frequency range.
SpectraLayers Pro 7's AI is amazing for breaking out stems, (as well as surgical repairs that cannot be accomplished in Sound Forge). It ain't perfect though and results vary depending on the material. It ISOs drums, vocals, keys and bass very well, but not very good for guitars and string instruments. It is a lot less costly than RX Advanced but far from low cost,