Yes, it's 100% client side, but you have to cache additional files. (A working set consists of the main script, a worker script for the application core and at least one voice definition to be used.)

Mind that the core won't run concurrently as a worker on mobile devices, but rather as an instance in the main/UI thread. This is, because mobile devices will mute the playback triggered by a message from a worker, as there is no immediate user interaction. Therefore, longer utterances are likely to block the UI noticeably, while the internal sound file is processed. This is a bit sad, but how things are.