podcast-splitter

播客分段器

根据静音检测自动将音频文件拆分为多个片段。非常适合将播客分成章节、从长录音中创建片段，或去除无声片段。

快速开始

from scripts.podcast_splitter import PodcastSplitter

# Auto-split by silence
splitter = PodcastSplitter("podcast_episode.mp3")
segments = splitter.split_by_silence()
splitter.export_segments("./chapters/")

# Remove long silences
splitter = PodcastSplitter("raw_recording.mp3")
splitter.remove_silence(min_length=2000)  # Remove silences > 2 seconds
splitter.save("clean_recording.mp3")

功能

• 静音检测：可配置阈值和时长
• 自动拆分：在自然停顿处拆分音频
• 去除静音：删除或缩短较长的停顿
• 导出章节：将各个片段保存为文件
• 预览模式：列出检测到的静音，不进行拆分
• 批量处理：处理多个文件

API 参考

初始化

splitter = PodcastSplitter("audio.mp3")

# With custom settings
splitter = PodcastSplitter(
    "audio.mp3",
    silence_thresh=-40,    # dBFS threshold
    min_silence_len=1000,  # Minimum silence length (ms)
    keep_silence=300       # Silence to keep at segment edges (ms)
)

静音检测

# Detect silence regions
silences = splitter.detect_silence()
# Returns: [(start_ms, end_ms), (start_ms, end_ms), ...]

# Print silence summary
splitter.print_silence_report()

拆分

# Split at all detected silences
segments = splitter.split_by_silence()

# Split at silences longer than threshold
segments = splitter.split_by_silence(min_silence_len=3000)

# Limit number of segments
segments = splitter.split_by_silence(max_segments=10)

去除静音

# Remove silences longer than threshold
splitter.remove_silence(min_length=2000)  # Remove >2s silences

# Shorten silences to max length
splitter.shorten_silence(max_length=500)  # Cap at 500ms

# Remove leading/trailing silence only
splitter.strip_silence()

导出

# Export all segments
splitter.export_segments(
    output_dir="./chapters/",
    prefix="chapter",       # chapter_01.mp3, chapter_02.mp3
    format="mp3",
    bitrate=192
)

# Export specific segments
splitter.export_segment(0, "intro.mp3")
splitter.export_segment(3, "conclusion.mp3")

# Save modified audio
splitter.save("output.mp3")

CLI 用法

# Split podcast into chapters
python podcast_splitter.py --input episode.mp3 --output-dir ./chapters/

# Detect and list silences (no splitting)
python podcast_splitter.py --input episode.mp3 --detect-only

# Remove long silences
python podcast_splitter.py --input raw.mp3 --output clean.mp3 --remove-silence 2000

# Custom sensitivity
python podcast_splitter.py --input episode.mp3 --output-dir ./chapters/ \
    --threshold -35 --min-silence 2000 --keep-silence 500

CLI 参数

参数	描述	默认值
--input	输入音频文件	必填
--output	输出文件（用于去除静音）	-
--output-dir	片段的输出目录	-
--detect-only	仅检测并报告静音	False
--threshold	静音阈值（dBFS）	-40
--min-silence	检测静音的最短时长（ms）	1000
--keep-silence	边缘保留的静音时长（ms）	300
--max-segments	要创建的最大片段数	None
--remove-silence	去除超过此时长的静音（ms）	-
--shorten-silence	将静音时长限制为（ms）	-
--prefix	输出文件名前缀	segment
--format	输出格式	mp3
--bitrate	输出比特率（kbps）	192

示例

将访谈拆分为问答片段

splitter = PodcastSplitter(
    "interview.mp3",
    silence_thresh=-35,     # Less sensitive (louder threshold)
    min_silence_len=2000,   # Only split on 2+ second pauses
    keep_silence=400        # Keep some silence for natural feel
)

segments = splitter.split_by_silence()
print(f"Found {len(segments)} segments")

splitter.export_segments("./questions/", prefix="qa")

去除录音中的无声片段

splitter = PodcastSplitter("raw_recording.mp3")

# Show what would be removed
splitter.print_silence_report()

# Remove silences longer than 3 seconds
splitter.remove_silence(min_length=3000)

# Cap remaining silences at 1 second
splitter.shorten_silence(max_length=1000)

splitter.save("clean_recording.mp3")

创建精彩片段

splitter = PodcastSplitter("episode.mp3")
segments = splitter.split_by_silence(min_silence_len=5000)

# Export only segments longer than 30 seconds
for i, segment in enumerate(segments):
    duration = segment['end'] - segment['start']
    if duration > 30000:  # > 30 seconds
        splitter.export_segment(i, f"highlight_{i+1}.mp3")

批量处理单集

import os
from scripts.podcast_splitter import PodcastSplitter

episodes_dir = "./raw_episodes/"
output_dir = "./processed/"

for filename in os.listdir(episodes_dir):
    if filename.endswith('.mp3'):
        filepath = os.path.join(episodes_dir, filename)
        splitter = PodcastSplitter(filepath)

        # Remove long silences
        splitter.remove_silence(min_length=2000)

        # Save cleaned version
        output_path = os.path.join(output_dir, filename)
        splitter.save(output_path)
        print(f"Processed: {filename}")

预览静音检测

splitter = PodcastSplitter("episode.mp3")

# Get detailed silence info
silences = splitter.detect_silence()

print("Detected Silences:")
for i, (start, end) in enumerate(silences):
    duration = (end - start) / 1000
    start_time = start / 1000
    print(f"  {i+1}. {start_time:.1f}s - {duration:.1f}s silence")

# Print summary
splitter.print_silence_report()

检测设置指南

音频类型	阈值	最短静音时长	备注
安静的录音室	-50 dBFS	500ms	灵敏度极高
普通播客	-40 dBFS	1000ms	默认值
嘈杂的录音	-35 dBFS	1500ms	灵敏度较低
有间奏的音乐	-30 dBFS	2000ms	适用于口播间隙

调整灵敏度

• 拆分更多：降低阈值（例如，-50），缩短 min_silence
• 拆分更少：提高阈值（例如，-30），延长 min_silence
• 听感自然：延长 keep_silence（500-1000ms）
• 紧凑剪辑：缩短 keep_silence（100-200ms）

依赖项

pydub>=0.25.0

注意：需要在系统上安装 FFmpeg。

限制

• 最适合语音内容（不适用于音乐）
• 嘈杂的录音可能需要调整阈值
• 长文件会占用大量内存
• 不会自动命名章节（需要手动重命名）