DubSplitter 0.7.0

an easy tool to split dubs based on given silence

DubSplitter

Description

an easy tool to split dubs based on given silence

Params

Command	Type	Info
-f, --fileName	option	file to process
-o, --outFilePath	option	output folder, if not set, will use scriptPath + \\Out\\ (as script), or userPath + \\DubSplitter\\Out\\ (as package)
--outFileFormat	option	output format, default is ogg
--fileNameFormat	option	output file name format
--fileNameVRFormat	option	output file name format with voice recognition
--fileNameCustomInfo	option	custom info for output file name, default is ''
-s, --silence	option	silence time, in ms, default is 1000ms
-r, --range	option	range, default is 100ms. e.g., silence = 400, range = 100 will slice in 400ms and 500ms
--step	option	loop step, default is 100ms
--threshold	option	anything quieter than this will be considered silence, default is -40db
--keepSilence	option	leave some silence at the beginning and end of the chunks. Keeps the sound from sounding like it is abruptly cut off. When the length of the silence is less than the given duration it is split evenly between the preceding and following non-silent segments, default is 100ms
--noVR	option	don't use voice recognition, default is false
--model	option	whisper model, default is base
--prompt	option	init prompt used in whisper, default is 简体中文
--language	option	language used in whisper, default is chinese
--omitLen	option	recognize result will omit middle characters if longer than given, len <=0 -> do nothing, default is 20
--log	option	output detailed log instead of progress bar, default is false

Usage

open folder in terminal, then run python main.py

or use command pip install DubSplitter to install package, then run dubSplitter

Custom File Name

Basic

fileNameFormat & fileNameVRFormat receives a format string, you can reference the formatting syntax doc then write your own one.

files will firstly be outputted in the format of fileNameFormat. If the script needs to do voice recognition, then the file will be renamed to fileNameVRFormat

File Name Format

default is {2:0>4d}_{3:0>8d}.{1}

String	Index
custom info	0
output format	1
silence	2
loop index	3
ms time stamp start	4
ms time stamp end	5
time stamp start	6
time stamp end	7

custom info is the one you passed in fileNameCustomInfo

File Name Format (with voice recognition)

default is {2:0>4d}_{3:0>8d}_{5}.{1}

String	Index
custom info	0
output format	1
silence	2
loop index	3
recognize_result	4
text	5
ms time stamp start	6
ms time stamp end	7
time stamp start	8
time stamp end	9

custom info is the one you passed in fileNameCustomInfo

text is the process result of recognize_result, by omitting middle characters, and escaping invalid characters like \\, /, *, ?, <, >, |

Note

Whisper GPU

if whisper doesn't use GPU, you need to uninstall CPU version first then install GPU version

pip uninstall torch
pip cache purge
# from https://pytorch.org/get-started/locally/
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

Whisper prompt

document

example:

use --prompt 简体中文 -> 真辛苦真辛苦啊 我会跳个好天气出去运动的

use --prompt 正體中文 -> 真辛苦真辛苦啊我會跳個好天氣出去運動的

Changelog

231102 0.7.0

• time stamp usable in custom file format

231027 0.6.0

• progress bar
• auto skip output routine if new silence has the same count comparing to previous one

231022 0.5.0

• iterate all files if input path is a folder

230520 0.4.0

• set silence threshold & keep silence length

230412 0.3.0

• add color for outputs
• add custom file format support
• add custom filename format support
• add prompt support
• add omit len option

230407 0.2.1

• use AudioSegment.from_file to support more file type
• load file before load model as file error happens more often
• remove unnecessary info & fix typo

230407 0.2.0

• print version when boot

230407 0.1.3

• fix typo

230407 0.1.2

• optimize update_path

230407 0.1.1

• update readme

230407 0.1.0

• init release