starmen1265
a350693863
- Merge short segments into paragraphs - Split on time gaps (>3 seconds by default) - Enforce max paragraph length (500 chars by default) - Simple rule-based approach (no LLM needed) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
98 lines
2.9 KiB
Python
98 lines
2.9 KiB
Python
"""Semantic Segmenter Module
|
|
|
|
将字幕分段按语义合并为段落。基于简单启发式规则：
|
|
1. 时间间隔超过阈值时分段
|
|
2. 段落长度限制
|
|
|
|
不使用 LLM 进行分段判断，保持简单和快速。
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from typing import List
|
|
|
|
from kb_workflow.srt.parser import Segment
|
|
|
|
|
|
class SemanticSegmenter:
|
|
"""语义分段器
|
|
|
|
将字幕分段转换为语义段落。基于时间和长度规则：
|
|
- 长时间停顿（>gap_threshold 秒）作为分段点
|
|
- 段落长度限制（max_paragraph_chars 字符）
|
|
|
|
Attributes:
|
|
gap_threshold: 时间间隔阈值（秒），默认 3.0
|
|
max_paragraph_chars: 单段落最大字符数，默认 500
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
gap_threshold: float = 3.0,
|
|
max_paragraph_chars: int = 500
|
|
):
|
|
"""初始化分段器
|
|
|
|
Args:
|
|
gap_threshold: 时间间隔阈值（秒），超过此值时分段
|
|
max_paragraph_chars: 单段落最大字符数
|
|
"""
|
|
self.gap_threshold = gap_threshold
|
|
self.max_paragraph_chars = max_paragraph_chars
|
|
|
|
def segment(self, segments: List[Segment]) -> List[str]:
|
|
"""将字幕分段转换为语义段落
|
|
|
|
Args:
|
|
segments: 字幕分段列表
|
|
|
|
Returns:
|
|
段落文本列表（已去除时间戳）
|
|
|
|
Example:
|
|
>>> segmenter = SemanticSegmenter()
|
|
>>> segments = [Segment(1, 0, 5, "第一句"), Segment(2, 5, 10, "第二句")]
|
|
>>> paragraphs = segmenter.segment(segments)
|
|
>>> len(paragraphs)
|
|
1
|
|
"""
|
|
if not segments:
|
|
return []
|
|
|
|
paragraphs = []
|
|
current_paragraph = []
|
|
current_length = 0
|
|
last_end_time = None
|
|
|
|
for seg in segments:
|
|
# 检查是否需要分段
|
|
should_start_new = False
|
|
|
|
# 规则1: 时间间隔超过阈值
|
|
if last_end_time is not None:
|
|
gap = seg.start_time - last_end_time
|
|
if gap > self.gap_threshold:
|
|
should_start_new = True
|
|
|
|
# 规则2: 段落长度超过限制
|
|
if current_length + len(seg.text) > self.max_paragraph_chars:
|
|
if current_paragraph: # 只有当前段落有内容时才分段
|
|
should_start_new = True
|
|
|
|
# 执行分段
|
|
if should_start_new and current_paragraph:
|
|
paragraphs.append(' '.join(current_paragraph))
|
|
current_paragraph = []
|
|
current_length = 0
|
|
|
|
# 添加当前分段
|
|
current_paragraph.append(seg.text)
|
|
current_length += len(seg.text)
|
|
last_end_time = seg.end_time
|
|
|
|
# 添加最后一个段落
|
|
if current_paragraph:
|
|
paragraphs.append(' '.join(current_paragraph))
|
|
|
|
return paragraphs
|