Speak While You Think: Streaming Speech Synthesis During Text Generation