[Request/Suggestion] Visualization of alignments
Opened this issue · 5 comments
I would like to request a feature.
It's nice to be able to easily align various audio files with adalign but it would also be nice to see how the files differ. When you explained the structure of the match dictionary to me I posted a screenshot of an offset graph. Something like this or a graph of wave forms with colored parts that match would be nice. I hope you get what I mean :D
Could you please incorporate something like this into audalign?
That's an interesting idea! I'm not entirely sure how that would work, though. Would this feature be included when plot=true is set in the config or the audalign.plot() function is called? Or, would it be a separate function?
Can you share the code you used for the offset graphs or make a pr?
A graph of color-coded waveforms seems neat, too, but it seems like it would be tough to set defaults. Maybe that would be a lot easier to accomplish in a GUI application?
Sorry, I have not really thought about where or how to incorporate this into audalign 😅. I think a separate function would be OK but having this as an extra graph in audalign.plot() would work too I guess.
Here is the code I wrote back then. I used this as a way to learn how to code a GUI application in Python but I'm not feeling very confident about the outcome.
I fingerprinted two files and sorted the outcome by the x_val (time) in def fingerprint In line 135 I created the plot. It uses a cursor (mplcursors) to show the actual time and offset.
import PySimpleGUI as sg
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
import audalign
import os
import pandas as pd
import mplcursors
import threading
import numpy as np
from itertools import groupby
# initialize audalign
ada = audalign.Audalign()
def fingerprint(orignal_filename, new_filename, window):
# fingerprint files
#ada = audalign.Audalign()
ada.fingerprint_file(orignal_filename)
ada.fingerprint_file(new_filename)
# recognize new file from fingerprints
result = ada.recognize(orignal_fingerprint_file, filter_matches=50, locality=5, locality_filter_prop = 0.1)
_,recognisefilename = os.path.split(new_filename)
if result != None:
x_val = []
y_val = []
# add up each offset found and sort by x_val (time)
for numberofoffsets in range(0,len(result['match_info'][recognisefilename]['offset_seconds'])):
for locality_seconds in result['match_info'][recognisefilename]['locality_seconds'][numberofoffsets]:
x_val.append(locality_seconds[0])
y_val.append(result['match_info'][recognisefilename]['offset_seconds'][numberofoffsets])
temp = sorted(zip(x_val, y_val))
x_val, y_val = zip(*temp)
# convert x_val to datetime
x_val = pd.to_datetime(x_val, unit='s')
window.write_event_value('-FINGERPRINTING DONE-', (x_val, y_val))
else:
window.write_event_value('-FINGERPRINTING DONE-', (None, None))
def convert_to_audio(filename, threadname, window):
name, extension = os.path.splitext(filename)
ada.convert_audio_file(filename, name + ".mp3")
window.write_event_value('-CONVERTING DONE-', threadname)
if __name__ == '__main__':
# set window theme
sg.theme("Default1")
# create ui layout
layout = [[sg.Text('Choose orignal file:')],
[sg.In(size=(102, 1), key='-ORIGFILE-', visible=True, enable_events=True),
sg.FileBrowse('Browse', file_types=(('video file', '*.mp4 *.avi *.mkv *.flv *.mov'), ('audio file', '*.mp3 *.wav'),('all files', '*.*')),key='-ORIGFILE_BTN-')],
[sg.Text('Choose new file:')],
[sg.In(size=(102, 1),key='-NEWFILE-', visible=True, enable_events=True),
sg.FileBrowse('Browse', file_types=(('video file', '*.mp4 *.avi *.mkv *.flv *.mov'), ('audio file', '*.mp3 *.wav'),('all files', '*.*')),key='-NEWFILE_BTN-')],
[sg.Button('Go'), sg.Button('Clear')],
[sg.Text('⚫', text_color='grey', key="ORIGFILECONVERTTHREAD", pad=(0,0), font='Default 14'),sg.Text('⚫', text_color='grey', key="NEWFILECONVERTTHREAD", pad=(0,0), font='Default 14'),sg.Text('⚫', text_color='grey', key="FINGERPRINTTHREAD", pad=(0,0), font='Default 14')]
]
# window = sg.Window('Offset Graph', layout, finalize=True)
window = sg.Window('Offset Graph', layout)
orignal_fingerprint_file_converted = False
new_fingerprint_file_converted = False
orignal_fingerprint_file = ""
new_fingerprint_file = ""
# --------------------- EVENT LOOP ---------------------
while True:
event, values = window.read()
if event == sg.WIN_CLOSED or event == 'Exit':
break
if event == 'Go':
if values["-ORIGFILE-"] != "" and values["-NEWFILE-"] != "" and values["-ORIGFILE-"] != values["-NEWFILE-"]:
window['-ORIGFILE-'].update(disabled=True)
window['-ORIGFILE_BTN-'].update(disabled=True)
window['-NEWFILE-'].update(disabled=True)
window['-NEWFILE_BTN-'].update(disabled=True)
orignal_fingerprint_file = values["-ORIGFILE-"]
# convert original file if needed
original_filename, original_file_extension = os.path.splitext(values["-ORIGFILE-"])
if not original_file_extension == ".wav" and not original_file_extension == ".mp3":
print("Original file is not an audio file. Converting to mp3...")
window["ORIGFILECONVERTTHREAD"].update(text_color='red')
thread_file1 = threading.Thread(target=convert_to_audio, args=(orignal_fingerprint_file, "ORIGFILECONVERTTHREAD", window,), daemon=True)
thread_file1.start()
orignal_fingerprint_file = original_filename + ".mp3"
else:
window["ORIGFILECONVERTTHREAD"].update(text_color='green')
orignal_fingerprint_file_converted = True
print("Original file to fingerprint: " + orignal_fingerprint_file)
new_fingerprint_file = values["-NEWFILE-"]
# convert newfile file if needed
newfile_filename, newfile_file_extension = os.path.splitext(values["-NEWFILE-"])
if not newfile_file_extension == ".wav" and not newfile_file_extension == ".mp3":
print("New file is not an audio file. Converting to mp3...")
window["NEWFILECONVERTTHREAD"].update(text_color='red')
thread_file2 = threading.Thread(target=convert_to_audio, args=(new_fingerprint_file, "NEWFILECONVERTTHREAD", window,), daemon=True)
thread_file2.start()
new_fingerprint_file = newfile_filename + ".mp3"
else:
window["NEWFILECONVERTTHREAD"].update(text_color='green')
new_fingerprint_file_converted = True
print("New file to fingerprint: " + new_fingerprint_file)
# start fingerprinting if files are audio files
if orignal_fingerprint_file_converted == True and new_fingerprint_file_converted == True:
print("No conversion necessary")
# fingerprint and recognize file
window["FINGERPRINTTHREAD"].update(text_color='red')
thread_fingerprint = threading.Thread(target=fingerprint, args=(orignal_fingerprint_file, new_fingerprint_file, window,), daemon=True)
thread_fingerprint.start()
else:
sg.popup('Choose (different) files first!')
elif event == '-CONVERTING DONE-':
threadname = values[event]
window[threadname].update(text_color='green')
if threadname == "ORIGFILECONVERTTHREAD":
orignal_fingerprint_file_converted = True
if threadname == "NEWFILECONVERTTHREAD":
new_fingerprint_file_converted = True
if orignal_fingerprint_file_converted == True and new_fingerprint_file_converted == True:
# fingerprint and recognize file
window["FINGERPRINTTHREAD"].update(text_color='red')
thread_fingerprint = threading.Thread(target=fingerprint, args=(orignal_fingerprint_file, new_fingerprint_file, window,), daemon=True)
thread_fingerprint.start()
elif event == '-FINGERPRINTING DONE-':
x_val, y_val = values[event]
window["FINGERPRINTTHREAD"].update(text_color='green')
print("Fingerprinting done")
# plot the results if there are ane xy_vals
if len(x_val) > 0:
plt.style.use('bmh')
fig = plt.figure()
ax = fig.gca()
ax.set_title("Offsets graph")
ax.xaxis.set_major_formatter(mdates.DateFormatter("%H:%M:%S"))
ax.yaxis.set_major_formatter(matplotlib.ticker.FormatStrFormatter('%.3f'))
#temp = sorted(zip(x_val, y_val))
#x_val, y_val = zip(*temp)
zippedVals = zip(x_val, y_val)
groupedlist = []
last_offset = None
templist = []
for offset, timestamp in zippedVals:
if last_offset == None:
templist.append((offset, timestamp))
else:
if offset == last_offset:
templist.append((offset, timestamp))
else:
groupedlist.append(templist)
templist = []
templist.append((offset, timestamp))
last_offset = offset
for group in groupedlist:
x_val, y_val = zip(*group)
print("Group")
print(list(x_val))
print(list(y_val))
# ax.plot(list(x_val), list(y_val))
# y = np.array(y_val)
# x = np.array(x_val)
# for yv in np.unique(y):
# if yv != np.nan:
# idx = y == yv
# ax.plot(x[idx],y[idx]) #,marker='o',linestyle='-',color='b'
lines = plt.gca()
# lines = ax.plot(x_val, y_val)
fig.autofmt_xdate()
# add cursor to plot
cursor = mplcursors.cursor(lines, hover = True)
def show_annotation(sel):
_, yi = sel.target
i = sel.target.index
timestamp = x_val[int(round(i))]
sel.annotation.set_text("Time: " + timestamp.strftime("%H:%M:%S") + "\n" + "Offset: " + str('%.3f' % yi))
cursor.connect('add', show_annotation)
plt.ylabel("Offset [s]")
plt.xlabel("Time")
fig.canvas.manager.set_window_title('')
plt.tight_layout()
plt.show(block=False)
else:
sg.popup('No matches found!')
elif event == 'Clear':
# clear paths
window['-ORIGFILE-'].update(value='')
window['-NEWFILE-'].update(value='')
# enable gui elements again
window['-ORIGFILE-'].update(disabled=False)
window['-ORIGFILE_BTN-'].update(disabled=False)
window['-NEWFILE-'].update(disabled=False)
window['-NEWFILE_BTN-'].update(disabled=False)
All this is probably not efficient and looks complicated. I was just trying to get it to work and I'm not really into programming.
About the color-coded waveforms. I'm always in awe when I see such nice looking plots in an applications like you have in your readme gif but I think matplotlib alone could also work. Even though it's not that simple to handle. It seems like one can color parts of a line or add bars to a plot. An image search for "matplotlib waveforms colored parts" shows some nice plots 😁
Something like this or this but with two or more waveforms having similar parts marked with semitransparent bars would be the optimum I think.
I don't think I'm able to code such features so I just made this post hoping for the best. Thank you for having an open ear.
Thanks for the references and code! It looks pretty solid.
That seems like a really neat feature, so I'll try to get it added in.
Hi everyone, just asked @jhpark16 to evaluate possible collaborations...
Here's a screenshot of his Video Audio Sync for FFMPEG tool:
Hope that inspires.
OT
@stephenjolly of @bbc added AudAlign in Similar Projects section of their audio-offset-finder repo.