VGM-Tools/spctools/ripsamples.py

#!/usr/bin/env python3

# ripsamples.py -- a python script for mass extracting samples from SPC files.
# (C) 2018 neoadpcmextract.c (C) 2018 a dinosaur (zlib)

import os
import subprocess
import pathlib
import struct
import hashlib

# Directory constants.
SPCDIR = "./spc"
ITDIR  = "./it"
SMPDIR = "./sample"

# External programs used by this script.
SPC2IT = "spc2it"


class Sample:
	length  = 0
	loopBeg = 0
	loopEnd = 0
	rate    = 0

	data = None

def writesmp(smp, path):
	with open(path, "wb") as wav:

		# Make sure sample rate is nonzero.
		#TODO: figure out why this even happens...
		if smp.rate == 0:
			smp.rate = 32000
			#print(path + " may be corrupted...")

		writeLoop = True if smp.loopEnd > smp.loopBeg else False

		# Write RIFF chunk.
		wav.write(b"RIFF")
		# Size of entire file following
		riffSize = 104 if writeLoop else 36
		wav.write(struct.pack("<I", riffSize + smp.length * 2))
		wav.write(b"WAVE")

		# Write fmt chunk.
		wav.write(b"fmt ")
		wav.write(struct.pack("<I", 16))            # Subchunk size.
		wav.write(struct.pack("<H", 1))             # Audio format (uncompressed)
		wav.write(struct.pack("<H", 1))             # Channel count (mono)
		wav.write(struct.pack("<I", smp.rate))      # Samplerate
		wav.write(struct.pack("<I", smp.rate * 2 )) # Byte rate (16 bit mono)
		wav.write(struct.pack("<H", 2))             # Bytes per sample (16 bit mono)
		wav.write(struct.pack("<H", 16))            # Bits per sample (16 bit)

		# Write sampler chunk (if looped).
		if writeLoop:
			wav.write(b"smpl")
			wav.write(struct.pack("<I", 60))    # Chunk size (36 + loops * 24)
			wav.write(b"\x00\x00\x00\x00")      # Manufacturer
			wav.write(b"\x00\x00\x00\x00")      # Product
			wav.write(b"\x00\x00\x00\x00")      # Sample period
			wav.write(b"\x00\x00\x00\x00")      # MIDI unity note
			wav.write(b"\x00\x00\x00\x00")      # MIDI pitch fraction
			wav.write(b"\x00\x00\x00\x00")      # SMPTE format
			wav.write(b"\x00\x00\x00\x00")      # SMPTE offset
			wav.write(struct.pack("<I", 1))     # Loop count
			wav.write(struct.pack("<I", 24))    # Loop data length

			wav.write(struct.pack("<I", 0))           # Cue point ID (none)
			wav.write(struct.pack("<I", 0))           # Loop type (forward)
			wav.write(struct.pack("<I", smp.loopBeg)) # Loop start
			wav.write(struct.pack("<I", smp.loopEnd)) # Loop end
			wav.write(struct.pack("<I", 0))           # Fraction (none)
			wav.write(struct.pack("<I", 0))           # Loop count (infinite)

		# Write data chunk.
		wav.write(b"data")
		wav.write(struct.pack("<I", smp.length * 2))
		wav.write(smp.data)

def readsmp(f, ofs, idx):
	# List of assumptions made:
	# - Samples are 16 bit
	# - Samples are mono
	# - Samples are signed
	# - No compression
	# - No sustain loop
	# - Loops are forwards
	# - Global volume is ignored

	f.seek(ofs)
	if f.read(4) != b"IMPS": return None

	# Skip fname to flags & read.
	f.seek(ofs + 0x12)
	flags = int.from_bytes(f.read(1), byteorder="little", signed=False)

	# Read flag values.
	if not flags & 0b00000001: return None # Check sample data bit.
	loopBit = True if flags & 0b00010000 else False

	smp = Sample()

	# Read the rest of the header.
	f.seek(ofs + 0x30)
	smp.length = int.from_bytes(f.read(4), byteorder="little", signed=False)
	if loopBit:
		smp.loopBeg = int.from_bytes(f.read(4), byteorder="little", signed=False)
		smp.loopEnd = int.from_bytes(f.read(4), byteorder="little", signed=False)
	else:
		f.seek(8, 1) # Skip over.
		smp.loopBeg = 0
		smp.loopEnd = 0
	smp.rate = int.from_bytes(f.read(4), byteorder="little", signed=False)
	f.seek(8, 1) # Skip over sustain shit.

	# Read sample data.
	dataOfs = int.from_bytes(f.read(4), byteorder="little", signed=False)
	smp.data = f.read(smp.length * 2)

	# Compute hash of data.
	#FIXME: This actually generates a butt ton of collisions...
	# there's got to be a better way!
	h = hashlib.md5(struct.pack("<pII", smp.data, smp.loopBeg, smp.loopEnd))
	smp.hash = h.hexdigest()

	return smp

def readit(path, outpath):
	with open(path, "r+b") as f:

		# Don't bother scanning non IT files.
		if f.read(4) != b"IMPM": return

		#print("Song name: " + f.read(26).decode('utf-8'))

		# Need order list size and num instruments to know how far to skip.
		f.seek(0x20)
		ordNum = int.from_bytes(f.read(2), byteorder="little", signed=False)
		insNum = int.from_bytes(f.read(2), byteorder="little", signed=False)
		smpNum = int.from_bytes(f.read(2), byteorder="little", signed=False)
		patNum = int.from_bytes(f.read(2), byteorder="little", signed=False)

		# Cus spc2it has a tendency to produce corrupted files...
		if ordNum > 1024: return
		if smpNum > 4000: return
		if insNum > 256: return
		if patNum > 256: return

		smpOfsTable = 0xC0 + ordNum + insNum * 4

		for i in range(0, smpNum):
			f.seek(smpOfsTable + i * 4)
			smpOfs = int.from_bytes(f.read(4), byteorder="little", signed=False)
			smp = readsmp(f, smpOfs, i + 1)
			if smp != None:
				outwav = os.path.join(outpath, smp.hash + ".wav")
				if not os.path.isfile(outwav):
					pathlib.Path(outpath).mkdir(parents=True, exist_ok=True)
					writesmp(smp, outwav)

def scanit(srcPath, dstPath):
	for directory, subdirectories, files in os.walk(srcPath):
		for file in files:
			if file.endswith(".it"):
				path = os.path.join(directory, file)
				outpath = dstPath + path[len(srcPath):-len(file)]
				readit(path, outpath)

def scanspc(srcPath, dstPath):
	for directory, subdirectories, files in os.walk(srcPath):

		# Create output dir for each game.
		for sub in subdirectories:
			path = os.path.join(dstPath, sub)
			pathlib.Path(path).mkdir(parents=True, exist_ok=True)

		# Convert spc files.
		for file in files:
			if file.endswith(".spc"):
				# Don't convert files that have already been converted.
				itpath = os.path.join(dstPath + directory[len(srcPath):], file[:-3] + "it")
				if not os.path.isfile(itpath):
					path = os.path.join(directory, file)
					subprocess.call([SPC2IT, path])
					path = path[:-3] + "it"
					if os.path.isfile(path):
						os.rename(path, itpath)


# Actual main stuff.
scanspc(SPCDIR, ITDIR)
scanit(ITDIR, SMPDIR)