Text is All You Need: Personalizing ASR Models using Controllable Speech Synthesis