Should I do a PhD in Sound and Music Computing?

SMC sounds like a very interesting but somewhat narrow specialization. Since it's not likely that an expert will appear to answer this question, here are some comments:

One key question regarding career paths is how bad you would feel if they fell through. For example, suppose you got a Ph.D. in SMC, tried and failed to get an academic job in this area, and ended up getting a job in industry that was relevant but didn't really require the full Ph.D. Would you regret having started on this path, or would you be happy to have had a chance to study something exciting and to have developed expertise that might still serve you well in the future? If you would regret it, then you need to think very carefully about the job market, but if your personality could handle this situation, then this path could be a wise choice.

The big danger with unusual fields is that few people will be specifically looking to hire in this area. You may get lucky and find someone who is, but you may need to create your own opportunities. If you have an outgoing personality and are good at networking and making connections, then there will be less risk.

In academia, you'll run into two difficulties. One is that no CS program will need to have this area represented, so you'll have to make a stronger case for why you would be a great hire. The other problem is that if a school is open to SMC, then they may already have someone in this area, and it's a narrow enough field that making multiple hires could be a very tough sell. So you would be looking for the schools that are interested, but not so interested that they have already hired someone. Of course, it's far from impossible, but some other branches of CS may be a little easier.

As for switching areas, it can certainly be done. You may run into a little resistance, depending on what you were originally hired to do. (This could be a serious issue in industry, and even in academia your colleagues may be counting on you to teach the introductory course in your old field.)

If you are equally interested in and talented at algorithms and SMC, then it's probably a little safer to start with algorithms. However, if only one of them will make you happy and inspire you to do your best work, then that one would be the better choice.

Every PhD is necessarily specialized, but along the way you must obtain different depths of knowledge from different fields.

So rather than selling yourself as someone researching Sound and Music Computing, you should mold yourself as someone doing, for example, Machine Learning with a focus on Sound and Music applications. This way, when you have finished your PhD doing what you love, you will still have skills that some department will be willing to hire you for. If you know machine learning, then you'll be able to teach machine learning courses. You'll be able to adapt that knowledge to solve other problems that may have similarities at some abstract level to Sound and Music, perhaps because they involve temporal streams of almost repeating data points.

When publishing, you will need to try to publish in top quality, general conferences or journals, rather than publishing everything in smaller, Sound and Music specific events. Naturally, the small events may provide you with good feedback and exposure within your community, but the larger events are what count when people are looking at your CV.

I think that it is very important to study what you find interesting, but do not overspecialized yourself into a tiny niche.