DiffVoice: Text-to-Speech with Latent Diffusion