mirror of
https://github.com/iluvcapra/wavinfo.git
synced 2026-01-01 09:20:40 +00:00
Fixes and testing with files
This commit is contained in:
100
wavinfo/wave_reader.py
Normal file
100
wavinfo/wave_reader.py
Normal file
@@ -0,0 +1,100 @@
|
||||
import struct
|
||||
|
||||
class WavInfoReader():
|
||||
"""
|
||||
format : WAV format
|
||||
bext : The Broadcast-WAV extension as definied by EBU Tech 3285 v2 (2011)
|
||||
|
||||
"""
|
||||
|
||||
def __init__(self, path):
|
||||
with open(path, 'rb') as f:
|
||||
chunks = parse_chunk(f)
|
||||
|
||||
main_list = chunks.children
|
||||
f.seek(0)
|
||||
|
||||
self.fmt = self._get_format(main_list,f)
|
||||
self.info = self._get_bext(main_list,f)
|
||||
self.ixml = self._get_ixml(main_list,f)
|
||||
|
||||
|
||||
def _get_format(self,chunks,f):
|
||||
fmt_chunk = next(chunk for chunk in chunks if chunk.ident == b'fmt ')
|
||||
fmt_data = fmt_chunk.read_data(f)
|
||||
|
||||
# The format chunk is
|
||||
# audio_format U16
|
||||
# channel_count U16
|
||||
# sample_rate U32 Note an integer
|
||||
# byte_rate U32 == SampleRate * NumChannels * BitsPerSample/8
|
||||
# block_align U16 == NumChannels * BitsPerSample/8
|
||||
# bits_per_sampl U16
|
||||
unpacked = struct.unpack("<HHIIHH", fmt_data)
|
||||
|
||||
return WavInfoFormat(audio_format = unpacked[0],
|
||||
channel_count = unpacked[1],
|
||||
sample_rate = unpacked[2],
|
||||
byte_rate = unpacked[3],
|
||||
block_align = unpacked[4],
|
||||
bits_per_sample = unpacked[5]
|
||||
)
|
||||
|
||||
def _get_bext(self,chunks,f):
|
||||
|
||||
bext_chunk = next((chunk for chunk in chunks if chunk.ident == b'bext'),None)
|
||||
bext_data = bext_chunk.read_data(from_stream=f)
|
||||
|
||||
# description[256]
|
||||
# originator[32]
|
||||
# originatorref[32]
|
||||
# originatordate[10] "YYYY:MM:DD"
|
||||
# originatortime[8] "HH:MM:SS"
|
||||
# lowtimeref U32
|
||||
# hightimeref U32
|
||||
# version U16
|
||||
# umid[64]
|
||||
#
|
||||
# EBU 3285 fields
|
||||
# loudnessvalue S16 (in LUFS*100)
|
||||
# loudnessrange S16 (in LUFS*100)
|
||||
# maxtruepeak S16 (in dbTB*100)
|
||||
# maxmomentaryloudness S16 (LUFS*100)
|
||||
# maxshorttermloudness S16 (LUFS*100)
|
||||
# reserved[180]
|
||||
# codinghistory []
|
||||
print(len(bext_data))
|
||||
packstring = "<256s"+ "32s" + "32s" + "10s" + "8s" + "QH" + "64s" + "hhhhh" + "180s"
|
||||
|
||||
rest_starts = struct.calcsize(packstring)
|
||||
unpacked = struct.unpack(packstring, bext_data[:rest_starts])
|
||||
|
||||
|
||||
return WavBextFormat(description=unpacked[0].decode('ascii').rstrip(' \t\r\n\0'),
|
||||
originator = unpacked[1].decode('ascii').rstrip(' \t\r\n\0'),
|
||||
originator_ref = unpacked[2].decode('ascii').rstrip(' \t\r\n\0'),
|
||||
originator_date = unpacked[3].decode('ascii').rstrip(' \t\r\n\0'),
|
||||
originator_time = unpacked[4].decode('ascii').rstrip(' \t\r\n\0'),
|
||||
time_reference = unpacked[5],
|
||||
version = unpacked[6],
|
||||
umid = unpacked[7],
|
||||
loudness_value = unpacked[8],
|
||||
loudness_range = unpacked[9],
|
||||
max_true_peak = unpacked[10],
|
||||
max_momentary_loudness = unpacked[11],
|
||||
max_shortterm_loudness = unpacked[12],
|
||||
coding_history = bext_data[rest_starts:].decode('ascii').rstrip(' \t\r\n\0')
|
||||
)
|
||||
|
||||
def _get_ixml(self,chunks,f):
|
||||
|
||||
ixml_chunk = next((chunk for chunk in chunks if chunk.ident == b'iXML'),None)
|
||||
ixml_data = ixml_chunk.read_data(from_stream=f)
|
||||
ixml_string = ixml_data.decode('utf-8')
|
||||
|
||||
return WavIXMLFormat(ixml_string)
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user