We revisit the task of speaker attribution in narrative texts from the underexplored perspective of industrial-scale audiobook production, and define Automated Book Drawing (ABD) as a structured solution to meet its unique annotation challenges. Motivated by the high cost and inconsistency of manual annotation in voice acting workflows, we formalize ABD as a multi-class classification task grounded in discourse reasoning. To support systematic research, we introduce BD-11, a high-quality Chinese benchmark dataset comprising 58,818 annotated dialogues from 11 diverse web novels, with dialogues categorized into easy and hard cases based on attribution difficulty. We benchmark three representative paradigms: (1) a heuristic method using named entity proximity; (2) a BERT-based Machine Reading Comprehension (MRC) model that treats attribution as a span extraction task; and (3) a prompt-based framework using ChatGPT with zero-shot, few-shot, and chain-of-thought prompting. Experimental results show that the MRC model achieves the highest overall accuracy (F1: 89.32%, EM: 87.70%), while ChatGPT demonstrates strong robustness in long-context and ambiguous scenarios. Our work contributes a domain-specific formulation, a publicly available dataset, and a comparative evaluation across multiple modeling approaches, laying a foundation for future research in character-centric narrative understanding and audiobook-focused NLP applications. The BD-11 dataset has been released and is publicly accessible at https://github.com/KF666-github/BD-11 .