diff --git a/README.md b/README.md
index 291cd15604f26bd9dcfd47b1e19f87eca8e23606..793b0e5c15056a43ba456f54c0197e92b010e443 100644
--- a/README.md
+++ b/README.md
@@ -83,6 +83,12 @@ $ deactivate
Having a CUDA-capable GPU is optional, but can greatly reduce processing time in some situations.
+## Configuration
+
+Autokara comes with a default config file in `autokara/default.conf`.
+If you want to tweak some values (enable CUDA, for example), you should add them to a new config file in your personal config directory : `~/.config/autokara/autokara.conf`.
+This new file has priority over the default one, which is used only as fallback for unspecified values.
+
# Use
diff --git a/autokara/autokara.py b/autokara/autokara.py
index e5d9e21da7dc6c38ac0a331baa7ecdbb968fe7db..ffb7edace7fad76076009591ed2a71cc06cc01b6 100644
--- a/autokara/autokara.py
+++ b/autokara/autokara.py
@@ -4,6 +4,7 @@ import demucs.separate
import subprocess
import shlex
from pathlib import Path
+from configparser import ConfigParser
from .autosyl.assUtils import AssWriter, getSyls, getHeader
from .autosyl.segment import segment
@@ -12,25 +13,6 @@ from .autosyl.segment import segment
def main(opts=None):
- from g2p import __path__ as g2p_path
-
- HERE = Path(__file__).parent
- g2p_base = Path(g2p_path[0])
- if not Path.exists(g2p_base / "mappings/langs/rji"):
- print("No Romaji language mapping found, attempting first-time setup")
- mapping_dir = HERE / "g2p/mappings/langs/"
- mappings = glob.glob(f"{str(mapping_dir):s}/*")
- for map in mappings:
- subprocess.check_call(shlex.split(f'cp -r {map:s} {str(g2p_base):s}/mappings/langs/'))
- subprocess.check_call(shlex.split(f'g2p update'))
-
- if not Path.exists(g2p_base / "mappings/langs/rji"):
- print("ERROR : Failed to find language mapping")
- else:
- print("Setup successful")
-
-
-
parser = argparse.ArgumentParser(description='AutoKara - Automatic karaoke timing tool')
parser.add_argument("source_file", type=str, help="The video/audio file to time")
parser.add_argument("ass_file", type=str, help="The ASS file with lyrics to time")
@@ -44,24 +26,41 @@ def main(opts=None):
ass_file = args.ass_file
verbose = args.verbose
+ here = Path(__file__).parent
+
+ config = ConfigParser()
+ config.read([
+ str(here / "default.conf"), # Default config file
+ str(Path().home()/ ".config" / "autokara"/ "autokara.conf") # User config file
+ ])
+
+ media_dir = config['Media']['media_dir']
+ segment_config = {
+ 'model': config['Segment']['model'],
+ 'bdr': config['Segment'].getboolean('bdr'),
+ 'cuda': config['Segment'].getboolean('cuda'),
+ 'syl_delay': config['Segment'].getint('syl_delay')
+ }
+
+
if not args.vocals :
print("Extracting audio from video file...")
- Path("./media/audio").mkdir(parents=True, exist_ok=True)
+ Path(media_dir + "/audio").mkdir(parents=True, exist_ok=True)
basename = Path(args.source_file).stem
- audio_file = "media/audio/%s.wav" % basename
+ audio_file = f"{media_dir:s}/audio/{basename:s}.wav"
- subprocess.call(shlex.split('./extractWav.sh "%s" "%s"' % (args.source_file, audio_file)))
+ subprocess.call(shlex.split(f'{str(here)}/extractWav.sh "{args.source_file:s}" "{audio_file}"'))
- Path("./media/vocals").mkdir(parents=True, exist_ok=True)
- output_folder = "./media/vocals"
+ Path(f"{media_dir:s}/vocals").mkdir(parents=True, exist_ok=True)
+ output_folder = f"{media_dir:s}/vocals"
print("Isolating vocals...")
# Not working, don't know why
# demucs.separate.main(shlex.split('--two-stems vocals -o "%s" "%s"' % (output_folder, audio_file)))
- subprocess.call(shlex.split('demucs --two-stems vocals -o "%s" "%s"' % (output_folder, audio_file)))
+ subprocess.call(shlex.split(f'demucs --two-stems vocals -o "{output_folder:s}" "{audio_file:s}"'))
- vocals_file = "./media/vocals/htdemucs/%s/vocals.wav" % basename
+ vocals_file = f"{media_dir:s}/vocals/htdemucs/{basename:s}/vocals.wav"
else:
vocals_file = args.source_file
@@ -76,7 +75,7 @@ def main(opts=None):
if verbose:
print("Starting syl detection...")
- syls = segment(vocals_file, reference_syls=reference_syls, verbose=verbose, language=args.lang)
+ syls = segment(vocals_file, segment_config, reference_syls=reference_syls, verbose=verbose, language=args.lang)
print(syls)
print(line_meta)
diff --git a/autokara/autosyl/segment.py b/autokara/autosyl/segment.py
index 84a5ac9bb42f4e6e2354d747cc9c80402255eff2..46155073b8c10738a55de11981295909f38c91e9 100644
--- a/autokara/autosyl/segment.py
+++ b/autokara/autosyl/segment.py
@@ -12,17 +12,17 @@ from .LyricsAlignment.wrapper import align, preprocess_from_file
-def segment(songfile, reference_syls=None, syls_per_line=10, last_syl_dur=500, verbose=False, language="jp"):
+def segment(songfile, config, reference_syls=None,syls_per_line=10, last_syl_dur=500, verbose=False, language="jp"):
- delay = -4
+ delay = config['syl_delay']
backtrack = False
print(reference_syls)
- audio_file = songfile # pre-computed source-separated vocals; These models do not work with mixture input.
- word_file = None # example: jamendolyrics/lyrics/*.words.txt"; Set to None if you don't have it
- method = "MTL_BDR" # "Baseline", "MTL", "Baseline_BDR", "MTL_BDR"
- cuda=False # set True if you have access to a GPU
+ audio_file = songfile # pre-computed source-separated vocals;
+ word_file = None # example: jamendolyrics/lyrics/*.words.txt"; Set to None if you don't have it
+ method = config['model'] + ("_BDR" if config['bdr'] else "") # "Baseline", "MTL", "Baseline_BDR", "MTL_BDR"
+ cuda = config['cuda'] # set True if you have access to a GPU
checkpoint_folder = f"{str(Path(__file__).parent):s}/LyricsAlignment/checkpoints"
language = language
diff --git a/autokara/default.conf b/autokara/default.conf
new file mode 100644
index 0000000000000000000000000000000000000000..82d914ce4a69e2792135de3cb620ee67f36e604e
--- /dev/null
+++ b/autokara/default.conf
@@ -0,0 +1,20 @@
+
+[General]
+
+
+
+[Media]
+
+# Where to store temporary media files (such as extracted vocals or ASS tracks)
+media_dir = /tmp/autokara
+
+[Segment]
+
+# Which model to use. Options are "Baseline" (base) and "MTL" (better)
+model = MTL
+# Whether to use BDR model for boundary recognition. Better, but resource-intensive
+bdr = true
+# Whether to use CUDA
+cuda = false
+# Default delay applied to detected syls, in centiseconds
+syl_delay = -4
diff --git a/preprocess_media.sh b/preprocess_media.sh
index a88ae744d830cadc1fdf2122e5c545180eaa4fb4..ec05203bec0a6563ab7adc7298b08c0fac3d2c4a 100755
--- a/preprocess_media.sh
+++ b/preprocess_media.sh
@@ -1,6 +1,20 @@
-
-
-
+#!/bin/bash
+
+
+##########################################################################################################
+#
+# COMMAND : preprocess_media.sh
+#
+# AUTHOR : Sting
+#
+# DESCRIPTION : CLI tool to batch extract ASS lyrics and vocals from a video folder
+#
+# USE : ./preprocess_media.sh input_folder output_folder
+#
+# REQUIREMENTS : FFMPEG, Demucs, extractAss and extractWav
+#
+#
+##########################################################################################################
USAGE_MESSAGE="usage : $0 video_folder train_folder"
diff --git a/setup.py b/setup.py
index 6f9b33985caa38b4b5217591a26ac3df162efb1c..c71a27b851a0ddfc76ffbe0efda851c06db10851 100644
--- a/setup.py
+++ b/setup.py
@@ -53,6 +53,11 @@ setup(
'autokara-gen-lang=autokara.update_lang_db:main'
],
},
+ scripts=[
+ 'autokara/extractAss.sh',
+ 'autokara/extractWav.sh',
+ 'preprocess_media.sh'
+ ],
license='MIT License',
classifiers=[
# Trove classifiers