A comprehensive voice persona dataset for character consistency in voice synthesis, generated using advanced audio-language model Qwen2-Audio-7B, with a GPU-optimized pipeline