Python library to split and join mp3 files

I wrote a library (pydub) for pretty much this exact use case:

from pydub import AudioSegment

sound = AudioSegment.from_mp3("/path/to/file.mp3")

# len() and slicing are in milliseconds
halfway_point = len(sound) / 2
second_half = sound[halfway_point:]

# Concatenation is just adding
second_half_3_times = second_half + second_half + second_half

# writing mp3 files is a one liner
second_half_3_times.export("/path/to/new/file.mp3", format="mp3")

Adding a silent gap

If you'd like to add silence between parts of a sound:

two_sec_silence = AudioSegment.silent(duration=2000)
sound_with_gap = sound[:1000] + two_sec_silence + sound[1000:]

Have a look at the MP3 file structure on Wikipedia. Use binary read mode in python to edit the MP3 file. s = open(file_name, 'rb').read() will put the whole file into a string object representing the raw bytes in your file (e.g. \xeb\xfe\x80). You can then search and edit the string, addressing the byte offsets with indeces using brackets: s[n]. Finally, just do a binary write of the MP3 frames you want in your new file(s), appending the ID3 header to the set of frames that you want to make up each file.

Here is my attempt at splitting an MP3 using python without re-encoding. Not all varieties of MP3 files are supported and I would gladly welcome suggestions or improvements. The script is hard coded to split at 55 seconds but the code demonstrates the general principles.

from __future__ import print_function
import struct
import sys

#MP3 frames are not independent because of the byte reservoir. This script does not account for
#that in determining where to do the split.

def SplitMp3(fi, splitSec, out):

    #Constants for MP3
    bitrates = {0x0: "free", 0x1: 32, 0x2: 40, 0x3: 48, 0x4: 56, 0x5: 64, 0x6: 80, 0x7: 96, 0x8: 112,
        0x9: 128, 0xa: 160, 0xb: 192, 0xc: 224, 0xd: 256, 0xe: 320, 0xf: "bad"}
    freqrates = {0x0: 44100, 0x1: 48000, 0x2: 32000, 0x3: "reserved"}
    countMpegFrames = 0
    frameDuration = 0.026
    unrecognizedBytes = 0
    splitFrame = int(round(splitSec / frameDuration))

    while True:

        startPos = fi.tell()

        #Check for 3 byte headers
        id3Start = fi.read(3)
        if len(id3Start) == 3:

            if id3Start == b'TAG':
                print ("Found ID3 v1/1.1 header")
                fi.seek(startPos + 256)
                continue

            if id3Start == b'ID3':
                #Possibly a ID3v2 header
                majorVer, minorVer, flags, encSize = struct.unpack(">BBBI", fi.read(7))
                if majorVer != 0xFF and minorVer != 0xFF:
                    encSize1 = (encSize & 0x7f000000) >> 24
                    encSize2 = (encSize & 0x7f0000) >> 16
                    encSize3 = (encSize & 0x7f00) >> 8
                    encSize4 = (encSize & 0x7f)
                    if encSize1 < 0x80 and encSize2 < 0x80 and encSize3 < 0x80 and encSize4 < 0x80:
                        size = ((encSize & 0x7f000000) >> 3) + ((encSize & 0x7f0000) >> 2) + ((encSize & 0x7f00) >> 1) + (encSize & 0x7f)
                        unsync = (flags >> 7) & 0x1
                        extendedHeader = (flags >> 6) & 0x1
                        experimental = (flags >> 5) & 0x1
                        print ("Found ID3v2 header")
                        print ("version", majorVer, minorVer, unsync, extendedHeader, experimental)
                        print ("size", size)
                        #TODO extendedHeader not supported yet

                        fi.seek(startPos + 10 + size)
                        continue

        #Check for 4 byte headers
        fi.seek(startPos)
        headerRaw = fi.read(4)
        if len(headerRaw) == 4:
            headerWord = struct.unpack(">I", headerRaw)[0]

            #Check for MPEG-1 audio frame
            if headerWord & 0xfff00000 == 0xfff00000:
                print ("Possible MPEG-1 audio header", hex(headerWord))
                countMpegFrames += 1
                ver = (headerWord & 0xf0000) >> 16
                bitrateEnc = (headerWord & 0xf000) >> 12
                freqEnc = (headerWord & 0xf00) >> 8
                mode = (headerWord & 0xf0) >> 4
                cpy = (headerWord & 0xf)
                if ver & 0xe == 0xa and freqEnc != 0xf:
                    print ("Probably an MP3 frame")
                    bitrate = bitrates[bitrateEnc]
                    freq = freqrates[freqEnc >> 2]
                    padding = ((freqEnc >> 1) & 0x1) == 1
                    print ("bitrate", bitrate, "kbps")
                    print ("freq", freq, "Hz")
                    print ("padding", padding)
                    frameLen = int((144 * bitrate * 1000 / freq ) + padding)

                    #Copy frame to output
                    fi.seek(startPos)
                    frameData = fi.read(frameLen)
                    if countMpegFrames >= splitFrame:
                        out.write(frameData)

                    fi.seek(startPos + frameLen)
                    continue
                else:
                    raise RuntimeError("Unsupported format:", hex(ver), "header:", hex(headerWord))

        #If no header can be detected, move on to the next byte
        fi.seek(startPos)
        nextByteRaw = fi.read(1)
        if len(nextByteRaw) == 0:
            break #End of file
        unrecognizedBytes += 1

    print ("unrecognizedBytes", unrecognizedBytes)
    print ("countMpegFrames", countMpegFrames)
    print ("duration", countMpegFrames * frameDuration, "sec")

if __name__=="__main__":
    fi = open(sys.argv[1], "rb")
    out = open("test.mp3", "wb")
    SplitMp3(fi, 55.0, out)
    out.close()

Merging would be a similar case of extracting and appending frames from two separate MP3 input files.

Check out GStreamer and its Python wrapper Gst-Python.