From 308fc8cb1b0dd741a13c366ea582632b3fbded6a Mon Sep 17 00:00:00 2001 From: Hawk Lee Date: Sun, 15 Feb 2026 20:27:23 +0800 Subject: [PATCH] fix(stitcher): prevent cut_end from bleeding into next sentence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two changes: 1. _expand_to_midpoints: cap expansion at 150ms beyond speech boundary instead of going all the way to the midpoint between sentences. 2. _refine_cut_point for cut_end: change direction from 'after' to 'before' — find where speech actually ends, don't push further toward the next sentence. --- aiia_podcast_stitcher.py | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-) diff --git a/aiia_podcast_stitcher.py b/aiia_podcast_stitcher.py index c055a12..ee369ad 100644 --- a/aiia_podcast_stitcher.py +++ b/aiia_podcast_stitcher.py @@ -353,7 +353,9 @@ class AIIA_Podcast_Stitcher: return best_sample / sr def _expand_to_midpoints(self, boundaries: list, total_duration: float) -> list: - """将切割点扩展到相邻句子间隙的中点，避免截断尾音/吸气声。""" + """将切割点扩展到相邻句子间隙中，但限制最大扩展量以避免吃进下一句。""" + MAX_EXPAND = 0.15 # 最多从语音边界向外扩展 150ms + if len(boundaries) <= 1: if boundaries: boundaries[0]["cut_start"] = 0.0 @@ -364,16 +366,18 @@ class AIIA_Podcast_Stitcher: if i == 0: boundaries[i]["cut_start"] = 0.0 else: - # 与前一句的间隙中点 + # 与前一句的间隙中点，但不超过 MAX_EXPAND gap_mid = (boundaries[i - 1]["end"] + boundaries[i]["start"]) / 2 - boundaries[i]["cut_start"] = round(gap_mid, 3) + boundaries[i]["cut_start"] = round( + max(gap_mid, boundaries[i]["start"] - MAX_EXPAND), 3) if i == len(boundaries) - 1: boundaries[i]["cut_end"] = total_duration else: - # 与后一句的间隙中点 + # 与后一句的间隙中点，但不超过 MAX_EXPAND gap_mid = (boundaries[i]["end"] + boundaries[i + 1]["start"]) / 2 - boundaries[i]["cut_end"] = round(gap_mid, 3) + boundaries[i]["cut_end"] = round( + min(gap_mid, boundaries[i]["end"] + MAX_EXPAND), 3) return boundaries @@ -474,9 +478,9 @@ class AIIA_Podcast_Stitcher: cut_start = boundary.get("cut_start", boundary["start"]) cut_end = boundary.get("cut_end", boundary["end"]) - # 基于能量的边界微调：方向性搜索，确保切割点只远离语音 + # 基于能量的边界微调：找到语音实际结束/开始位置 cut_start = self._refine_cut_point(wav, sr, cut_start, direction="before") - cut_end = self._refine_cut_point(wav, sr, cut_end, direction="after") + cut_end = self._refine_cut_point(wav, sr, cut_end, direction="before") # 应用 padding cut_start = max(0, cut_start - padding)