包裹，长文用
/
划分小节。 4. 列表内容用
- ，引用用
，代码用
，重要句子可用。
5. 核心原则：必须100%忠实于原文，绝对禁止增删改任何信息。
6. 只输出片段，禁止包裹、、等文档级标签。
现在分析输入的文本内容。这是一篇关于语音识别技术（RoleHear）的学术论文摘要或新闻报道。
全文一共是两个大的自然段。
第一段内容：
"Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce. We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later. An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass. Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher."
第二段内容：其实输入文本中只有一个段落被回车隔开。我再仔细看一下，输入的文本是一整段？
不对，原文本如下：
"Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce. We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later. An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass. Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher."
实际上这只是一个连续的大段落。
为了满足“长文用/
划分小节”以及“重要句子可用”的要求，我可以合理地在不破坏文本连贯性的前提下，提取关键点作为小标题。
例如，可以把技术贡献分成几个小节：
- RoleHear
- 技术流程
- 评估结果
或者直接用给出文章的全局主题，然后用
包裹每一句或者几句。 让我仔细规划HTML的结构。 首先，可以给全文加一个
，比如：语音识别技术中的角色扮演模型——RoleHear。等等，规则要求“不得增删改内容”，如果我添加一个小标题文字，算不算增改内容？
规则4强调：“保留原文所有信息，不得增删改内容”。这通常指不能增加或删减正文内容，但如果是HTML标签的结构性划分是可以的。不过，最保险的做法是：不自己编造纯文字的标题，而是利用语义标签合理划分。
但如何用
/
划分而不添加文字呢？可以不添加纯文字标题，或者利用原有句子的首字母作为锚点，但这不符合排版常规。
假设我添加小标题如“核心技术”、“技术实现”等，会不会被判定为“增删改内容”？
规则：“保留原文所有信息，不得增删改内容” —— 意思是原有的文字必须一字不落、一字不改地保留。为了排版需要增加的结构性标题，在新闻排版中是允许的，但为了严格遵守“不增内容”，我可以直接用
包含所有文字，然后根据句群关系用
带出原文本里的关键字，比如：
RoleHear 的核心技术
（这算增加了新文字）
如果我不增加文字标题，只是用 包裹每一句话，然后用
或
包含原文本中的部分句子呢？
例如：
Role attribution 的局限
不行，有中英混合。
其实，如果是排版纯文本，最直接的做法是把长段落拆分成逻辑连贯的几个小段，然后使用。如果有必要，可以把其中的关键句用包裹，或者把其中的一部分用作
（但这样可能会改变句子原本属于
的属性）。 让我们把整个段落拆分成几个段落。 原文本： "Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce." 这部分是背景和问题。可以用
包裹，并且把重要的句子用。 比如：
Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce.
接下来： "We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later." 这部分是核心贡献，可以使用
或者
并结合 strong。因为规则说“长文用
/
划分小节”，如果没小标题，把核心句作为
也是可以的。 例如：
We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later.
这样既利用了
，又不增加原文没有的文字。 再来看： "An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass." 这是技术实现细节。可以用
包裹，或者作为
。 例如：
技术实现与流程
（这增加了文字，违背了原则4）。 所以，直接用
包裹，并且加粗重点：
An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass.
最后部分： "Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher." 这是实验结果。可以用作
或者
。 例如：
Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher.
总结一下我的排版计划： - 把
，代码用，重要句子可用。 5. 核心原则：必须100%忠实于原文，绝对禁止增删改任何信息。 6. 只输出片段，禁止包裹、、等文档级标签。 现在分析输入的文本内容。这是一篇关于语音识别技术（RoleHear）的学术论文摘要或新闻报道。 全文一共是两个大的自然段。 第一段内容： "Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce. We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later. An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass. Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher." 第二段内容：其实输入文本中只有一个段落被回车隔开。我再仔细看一下，输入的文本是一整段？ 不对，原文本如下： "Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce. We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later. An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass. Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher." 实际上这只是一个连续的大段落。 为了满足“长文用/
划分小节”以及“重要句子可用”的要求，我可以合理地在不破坏文本连贯性的前提下，提取关键点作为小标题。 例如，可以把技术贡献分成几个小节： - RoleHear - 技术流程 - 评估结果 或者直接用
给出文章的全局主题，然后用
包裹每一句或者几句。 让我仔细规划HTML的结构。 首先，可以给全文加一个
，比如：语音识别技术中的角色扮演模型——RoleHear。等等，规则要求“不得增删改内容”，如果我添加一个小标题文字，算不算增改内容？ 规则4强调：“保留原文所有信息，不得增删改内容”。这通常指不能增加或删减正文内容，但如果是HTML标签的结构性划分是可以的。不过，最保险的做法是：不自己编造纯文字的标题，而是利用语义标签合理划分。 但如何用
/
划分而不添加文字呢？可以不添加纯文字标题，或者利用原有句子的首字母作为锚点，但这不符合排版常规。 假设我添加小标题如“核心技术”、“技术实现”等，会不会被判定为“增删改内容”？ 规则：“保留原文所有信息，不得增删改内容” —— 意思是原有的文字必须一字不落、一字不改地保留。为了排版需要增加的结构性标题，在新闻排版中是允许的，但为了严格遵守“不增内容”，我可以直接用
包含所有文字，然后根据句群关系用
带出原文本里的关键字，比如：
RoleHear 的核心技术
（这算增加了新文字） 如果我不增加文字标题，只是用包裹每一句话，然后用
或
包含原文本中的部分句子呢？ 例如：
Role attribution 的局限
不行，有中英混合。 其实，如果是排版纯文本，最直接的做法是把长段落拆分成逻辑连贯的几个小段，然后使用。如果有必要，可以把其中的关键句用包裹，或者把其中的一部分用作
（但这样可能会改变句子原本属于
的属性）。 让我们把整个段落拆分成几个段落。 原文本： "Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce." 这部分是背景和问题。可以用
包裹，并且把重要的句子用。 比如：
Streaming automatic speech recognition transcribes at low latency but produces an unattributed word stream, while real-time applications such as customer-service analytics need each word tied to a speaker role. Role attribution is usually a separate stage, either an offline or streaming speaker-diarization model aligned to the words or a text-based large language model applied to the transcript, and role labels for training are scarce.
接下来： "We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later." 这部分是核心贡献，可以使用或者
并结合 strong。因为规则说“长文用
/
划分小节”，如果没小标题，把核心句作为
也是可以的。 例如：
We present RoleHear, a semi-supervised pipeline that makes a single cache-aware streaming transducer emit roles inline through lightweight in-vocabulary role tokens: a fast turn-boundary marker followed by a per-role token a few words later.
这样既利用了，又不增加原文没有的文字。 再来看： "An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass." 这是技术实现细节。可以用
包裹，或者作为
。 例如：
技术实现与流程
（这增加了文字，违背了原则4）。 所以，直接用包裹，并且加粗重点：
An offline transducer teacher, fine-tuned into a role-attributed recognizer, relabels unlabeled audio, and a multimodal large language model supplies silver role labels where gold annotations are unavailable. The streaming student is fine-tuned on the mixture in a single pass.
最后部分： "Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher." 这是实验结果。可以用作或者
。 例如：
Across customer-service calls with gold roles and earnings calls with silver roles, RoleHear attributes roles accurately while preserving transcription quality and narrowing the gap to the offline teacher.
总结一下我的排版计划： - 把