如何在Python中处理重叠音频字节流，仅写入非重叠部分？

核心思路是先从文本层面确定重叠片段，再通过音频时长对应截取字节流——直接对比音频字节不可靠，相同文本的TTS合成字节可能因语调、合成参数差异出现细微不同。具体步骤如下：

1. 定位文本的非重叠部分 #

先找出当前句子与前一句的重叠前缀，提取出当前句的非重叠后缀，用最长公共子串匹配实现：

def find_non_overlap(prev_text, curr_text):
    max_overlap = 0
    # 遍历可能的重叠长度，找到最长匹配
    for k in range(1, min(len(prev_text), len(curr_text)) + 1):
        if curr_text.startswith(prev_text[-k:]):
            max_overlap = k
    return curr_text[max_overlap:]

2. 计算重叠文本对应的音频时长 #

把重叠文本单独调用TTS API生成音频，用pydub解析时长。如果你的API支持直接返回音频时长元数据，可跳过此步骤直接用API返回值：

from pydub import AudioSegment
import io

def get_overlap_duration(overlap_text):
    # 生成重叠文本的音频流
    overlap_stream = API_call(overlap_text)
    overlap_bytes = overlap_stream.read()
    # 加载音频并获取时长（秒）
    audio = AudioSegment.from_file(io.BytesIO(overlap_bytes), format="mp3")
    return audio.duration_seconds

3. 截取音频流的非重叠部分并追加 #

加载当前完整音频流，跳过重叠时长对应的片段，只保留非重叠部分追加到目标文件：

from pydub import AudioSegment
import io

prev_text = ""  # 保存前一句文本，而非音频流（音频流是一次性的）

for curr_text in sentences:
    curr_stream = API_call(curr_text)
    curr_bytes = curr_stream.read()
    
    if prev_text:
        # 确认当前文本的前缀是否与前文本的后缀重叠
        max_possible_overlap = min(len(prev_text), len(curr_text))
        overlap_text = prev_text[-max_possible_overlap:]
        if curr_text.startswith(overlap_text):
            # 获取重叠部分的时长，截取非重叠音频
            overlap_duration = get_overlap_duration(overlap_text)
            full_audio = AudioSegment.from_file(io.BytesIO(curr_bytes), format="mp3")
            non_overlap_audio = full_audio[overlap_duration * 1000:]  # pydub用毫秒计时
            
            # 追加到目标文件
            with open("file.mp3", "ab") as f:
                non_overlap_audio.export(f, format="mp3")
        else:
            # 无重叠，直接写入完整音频
            with open("file.mp3", "ab") as f:
                f.write(curr_bytes)
    else:
        # 第一句直接全量写入
        with open("file.mp3", "ab") as f:
            f.write(curr_bytes)
    
    prev_text = curr_text

注意事项 #

• 需安装依赖：pip install pydub，同时要安装ffmpeg（pydub依赖它处理音频编解码）。
• 如果TTS API返回的音频格式不是MP3，需修改代码中的format参数为对应格式（如wav）。
• 若当前句完全是前一句的重叠，截取后的音频会是空的，此时不写入任何内容即可。

内容的提问来源于stack exchange，提问作者skidjoe