ahmed/acousticbrainz-ng
1
0
Fork 0
Code Issues Pull Requests Releases Activity

Run binaries on song

Browse Source
This commit is contained in:
Ahmed Al-Taiar
2025-08-06 15:38:22 -04:00
commit 9c0b8f9ab5
35 changed files with 1788 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored Normal file
View File

@@ -0,0 +1,2 @@
__pycache__/
cache/

493
__init__.py Normal file
View File

@@ -0,0 +1,493 @@
import os
import subprocess
import hashlib
from picard import config, log
from picard.ui.options import (
OptionsPage,
register_options_page,
)
from picard.ui.itemviews import (
BaseAction,
register_track_action,
# register_album_action,
)
from picard.track import Track
from PyQt5 import QtWidgets
import threading
import concurrent.futures
PLUGIN_NAME = "AcousticBrainz-ng"
PLUGIN_AUTHOR = "cy1der"
PLUGIN_DESCRIPTION = """
Analyze track acoustic characteristics using Essentia
<br/>
This plugin is not affiliated with the <a href='https://acousticbrainz.org'>AcousticBrainz</a> project<br/>
This is not a 1:1 recreation of the AcousticBrainz schema, but will provide most of the meaningful data<br/>
External dependencies:
<ul>
<li><a href='https://essentia.upf.edu'>Essentia</a> binaries compiled with TensorFlow and gaia2 support</li>
<li>A few MusicNN models (see user guide for details)</li>
</ul>
<strong>This plugin is CPU heavy!</strong>
"""
PLUGIN_VERSION = "1.0.0"
PLUGIN_API_VERSIONS = ["2.7", "2.8", "2.9", "2.10", "2.11"]
PLUGIN_LICENSE = "GPL-2.0-or-later"
PLUGIN_LICENSE_URL = "https://www.gnu.org/licenses/gpl-2.0.html"
PLUGIN_USER_GUIDE_URL = "https://example.com" # TODO: Update with actual user guide URL
REQUIRED_MODELS: list[tuple[str, str]] = [
("msd-musicnn-1", "msd.json"),
("mood_acoustic-musicnn-mtt-2", "mood_acoustic.json"),
("mood_aggressive-musicnn-mtt-2", "mood_aggressive.json"),
("mood_electronic-musicnn-msd-2", "mood_electronic.json"),
("mood_happy-musicnn-msd-2", "mood_happy.json"),
("mood_party-musicnn-mtt-2", "mood_party.json"),
("mood_relaxed-musicnn-msd-2", "mood_relaxed.json"),
("mood_sad-musicnn-msd-2", "mood_sad.json"),
("danceability-musicnn-msd-2", "danceability.json"),
("gender-musicnn-msd-2", "gender.json"),
("tonal_atonal-musicnn-mtt-2", "tonality.json"),
("voice_instrumental-musicnn-msd-2", "voice_instrumental.json")
]
OPTIONAL_MODELS: list[tuple[str, str]] = [
("genre_electronic-musicnn-msd-2", "genre_electronic.json"),
("genre_rosamerica-musicnn-msd-2", "genre_rosamerica.json"),
("genre_tzanetakis-musicnn-msd-2", "genre_tzanetakis.json")
]
REQUIRED_BINARIES: list[str] = [
"streaming_extractor_music",
"streaming_musicnn_predict",
"streaming_md5",
]
# Avoid memory hogging
TF_ENABLE_ONEDNN_OPTS: int = 0
ENV = os.environ.copy()
ENV['TF_ENABLE_ONEDNN_OPTS'] = str(TF_ENABLE_ONEDNN_OPTS)
config.TextOption("setting", "acousticbrainz_ng_binaries_path", os.path.join(os.path.dirname(__file__), "bin"))
config.TextOption("setting", "acousticbrainz_ng_models_path", os.path.join(os.path.dirname(__file__), "models"))
config.TextOption("setting", "acousticbrainz_ng_cache_path", os.path.join(os.path.dirname(__file__), "cache"))
config.IntOption("setting", "acousticbrainz_ng_max_musicnn_workers", 4)
config.BoolOption("setting", "acousticbrainz_ng_autorun", False)
config.BoolOption("setting", "acousticbrainz_ng_analyze_optional", False)
config.BoolOption("setting", "acousticbrainz_ng_save_raw", False)
class AcousticBrainzNGOptionsPage(OptionsPage):
NAME = "acousticbrainz_ng"
TITLE = "AcousticBrainz-ng"
PARENT = "plugins"
def __init__(self, parent=None) -> None:
super().__init__(parent)
self.setup_ui()
def _create_path_input_layout(self, line_edit: QtWidgets.QLineEdit, browse_callback, check_callback=None) -> QtWidgets.QHBoxLayout:
layout = QtWidgets.QHBoxLayout()
browse_button = QtWidgets.QPushButton("Browse", self)
browse_button.clicked.connect(browse_callback)
layout.addWidget(line_edit)
layout.addWidget(browse_button)
if check_callback:
check_button = QtWidgets.QPushButton("Check", self)
check_button.clicked.connect(check_callback)
layout.addWidget(check_button)
return layout
def setup_ui(self) -> None:
layout = QtWidgets.QVBoxLayout(self)
options_group = QtWidgets.QGroupBox("Options", self)
options_group.setSizePolicy(QtWidgets.QSizePolicy.Preferred, QtWidgets.QSizePolicy.Minimum)
options_layout = QtWidgets.QVBoxLayout(options_group)
self.autorun_checkbox = QtWidgets.QCheckBox("Autorun analysis", self)
self.autorun_checkbox.setToolTip("Automatically run analysis on new tracks")
self.analyze_optional_checkbox = QtWidgets.QCheckBox("Analyze optional models", self)
self.analyze_optional_checkbox.setToolTip("Include optional models in the analysis")
self.save_raw_checkbox = QtWidgets.QCheckBox("Save raw values", self)
self.save_raw_checkbox.setToolTip("Save raw MusicNN numbers in the metadata")
musicnn_workers_layout = QtWidgets.QHBoxLayout()
musicnn_workers_label = QtWidgets.QLabel("Max MusicNN workers:", self)
musicnn_workers_label.setSizePolicy(QtWidgets.QSizePolicy.Minimum, QtWidgets.QSizePolicy.Preferred)
self.musicnn_workers_input = QtWidgets.QSpinBox(self)
self.musicnn_workers_input.setToolTip("Maximum number of concurrent MusicNN workers")
self.musicnn_workers_input.setRange(1, max(len(REQUIRED_MODELS), len(OPTIONAL_MODELS)))
self.musicnn_workers_input.setSizePolicy(QtWidgets.QSizePolicy.Minimum, QtWidgets.QSizePolicy.Preferred)
musicnn_workers_layout.addWidget(musicnn_workers_label)
musicnn_workers_layout.addStretch()
musicnn_workers_layout.addWidget(self.musicnn_workers_input)
options_layout.addWidget(self.autorun_checkbox)
options_layout.addWidget(self.analyze_optional_checkbox)
options_layout.addWidget(self.save_raw_checkbox)
options_layout.addLayout(musicnn_workers_layout)
layout.addWidget(options_group)
paths_group = QtWidgets.QGroupBox("Paths", self)
paths_group.setSizePolicy(QtWidgets.QSizePolicy.Preferred, QtWidgets.QSizePolicy.Minimum)
paths_layout = QtWidgets.QVBoxLayout(paths_group)
# Binaries path
self.binaries_path_input = QtWidgets.QLineEdit(self)
self.binaries_path_input.setPlaceholderText("Path to Essentia binaries")
binaries_layout = self._create_path_input_layout(
self.binaries_path_input,
lambda: self._browse_folder(self.binaries_path_input),
lambda: (self._check_binaries(show_success=True), None)[1]
)
# Models path
self.models_path_input = QtWidgets.QLineEdit(self)
self.models_path_input.setPlaceholderText("Path to MusicNN models")
models_layout = self._create_path_input_layout(
self.models_path_input,
lambda: self._browse_folder(self.models_path_input),
lambda: (self._check_models(show_success=True, check_optional=True), None)[1]
)
# Cache path
self.cache_path_input = QtWidgets.QLineEdit(self)
self.cache_path_input.setPlaceholderText("Path to cache directory")
cache_layout = self._create_path_input_layout(
self.cache_path_input,
lambda: self._browse_folder(self.cache_path_input)
)
paths_layout.addWidget(QtWidgets.QLabel("Binaries", self))
paths_layout.addLayout(binaries_layout)
paths_layout.addWidget(QtWidgets.QLabel("Models", self))
paths_layout.addLayout(models_layout)
paths_layout.addWidget(QtWidgets.QLabel("Cache", self))
paths_layout.addLayout(cache_layout)
layout.addWidget(paths_group)
layout.addStretch()
def _check_binaries(self, show_success=False) -> bool:
path = self.binaries_path_input.text()
if not path or not os.path.exists(path):
QtWidgets.QMessageBox.warning(self, "Binaries", "Invalid or empty path.")
return False
missing_binaries = []
for binary in REQUIRED_BINARIES:
binary_path = AcousticBrainzNG._get_binary_path(binary, path)
if not os.path.exists(binary_path):
missing_binaries.append(binary)
if missing_binaries:
message = f"Missing binaries:\n" + "\n".join(f"• {binary}" for binary in missing_binaries)
QtWidgets.QMessageBox.warning(self, "Binaries", message)
return False
else:
if show_success:
QtWidgets.QMessageBox.information(self, "Binaries", "All binaries found!")
return True
def _check_models(self, show_success=False, check_optional=True) -> bool:
path = self.models_path_input.text()
if not path or not os.path.exists(path):
QtWidgets.QMessageBox.warning(self, "Models", "Invalid or empty path.")
return False
missing_required = []
for model in REQUIRED_MODELS:
model_path = os.path.join(path, f"{model[0]}.pb")
if not os.path.exists(model_path):
missing_required.append(model[0])
missing_optional = []
if check_optional:
for model in OPTIONAL_MODELS:
model_path = os.path.join(path, f"{model[0]}.pb")
if not os.path.exists(model_path):
missing_optional.append(model[0])
if missing_required:
message = f"Missing required models:\n" + "\n".join(f"• {model}.pb" for model in missing_required)
QtWidgets.QMessageBox.warning(self, "Models", message)
return False
elif missing_optional and check_optional:
message = f"Missing optional models:\n" + "\n".join(f"• {model}.pb" for model in missing_optional)
QtWidgets.QMessageBox.information(self, "Models", message)
if show_success:
if missing_optional and check_optional:
QtWidgets.QMessageBox.information(self, "Models", "All required models found! Some optional models are missing.")
else:
QtWidgets.QMessageBox.information(self, "Models", "All models found!")
return True
def _browse_folder(self, line_edit: QtWidgets.QLineEdit) -> None:
folder = QtWidgets.QFileDialog.getExistingDirectory(
self, "Select Folder",
line_edit.text() or os.path.expanduser("~")
)
if folder:
line_edit.setText(folder)
def load(self):
self.autorun_checkbox.setChecked(config.setting["acousticbrainz_ng_autorun"] or False)
self.analyze_optional_checkbox.setChecked(config.setting["acousticbrainz_ng_analyze_optional"] or False)
self.save_raw_checkbox.setChecked(config.setting["acousticbrainz_ng_save_raw"] or False)
self.musicnn_workers_input.setValue(config.setting["acousticbrainz_ng_max_musicnn_workers"] or 4)
self.binaries_path_input.setText(config.setting["acousticbrainz_ng_binaries_path"])
self.models_path_input.setText(config.setting["acousticbrainz_ng_models_path"])
self.cache_path_input.setText(config.setting["acousticbrainz_ng_cache_path"])
def save(self):
self._check_binaries()
self._check_models(show_success=False, check_optional=False)
config.setting["acousticbrainz_ng_autorun"] = self.autorun_checkbox.isChecked()
config.setting["acousticbrainz_ng_analyze_optional"] = self.analyze_optional_checkbox.isChecked()
config.setting["acousticbrainz_ng_save_raw"] = self.save_raw_checkbox.isChecked()
max_workers = max(1, min(self.musicnn_workers_input.value(), max(len(REQUIRED_MODELS), len(OPTIONAL_MODELS))))
config.setting["acousticbrainz_ng_max_musicnn_workers"] = max_workers
config.setting["acousticbrainz_ng_binaries_path"] = self.binaries_path_input.text()
config.setting["acousticbrainz_ng_models_path"] = self.models_path_input.text()
config.setting["acousticbrainz_ng_cache_path"] = self.cache_path_input.text()
class AcousticBrainzNG:
def __init__(self):
pass
@staticmethod
def _get_binary_path(binary_name: str, binaries_path: str) -> str:
binary_path = os.path.join(binaries_path, binary_name)
if os.name == 'nt': # Windows
binary_path += '.exe'
return binary_path
def _get_binary_paths(self) -> tuple[str, str]:
binaries_path = config.setting["acousticbrainz_ng_binaries_path"]
if not binaries_path:
raise ValueError("Binaries path not configured")
musicnn_binary_path = self._get_binary_path("streaming_musicnn_predict", binaries_path)
gaia_binary_path = self._get_binary_path("streaming_extractor_music", binaries_path)
if not os.path.exists(musicnn_binary_path):
raise FileNotFoundError(f"Binary {musicnn_binary_path} not found")
if not os.path.exists(gaia_binary_path):
raise FileNotFoundError(f"Binary {gaia_binary_path} not found")
return musicnn_binary_path, gaia_binary_path
def _run_musicnn_models(self, models: list[tuple[str, str]], musicnn_binary_path: str, file: str, output_path: str) -> None:
models_path = config.setting["acousticbrainz_ng_models_path"]
if not models_path:
raise ValueError("Models path not configured")
def run_musicnn_model(model_info):
model_name, output_file = model_info
try:
model_path = os.path.join(models_path, f"{model_name}.pb")
if not os.path.exists(model_path):
raise FileNotFoundError(f"Model {model_name} not found at {model_path}")
output_file_path = os.path.join(output_path, output_file)
if os.path.exists(output_file_path):
log.debug(f"{output_file_path} already exists, skipping {model_name}")
return
subprocess.run(
[musicnn_binary_path, model_path, file, output_file_path],
capture_output=True,
text=True,
env=ENV
)
except FileNotFoundError as e:
log.error(f"Model {model_name} not found: {e}")
except Exception as e:
log.error(f"Error processing model {model_name}: {e}")
max_workers = config.setting["acousticbrainz_ng_max_musicnn_workers"] or 4
with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
futures = [executor.submit(run_musicnn_model, model) for model in models]
concurrent.futures.wait(futures)
def analyze_required(self, metadata: dict, file: str) -> None:
if not self._check_binaries():
log.error("Essentia binaries not found")
return
if not self._check_required_models():
log.error("Required models not found")
return
try:
musicnn_binary_path, gaia_binary_path = self._get_binary_paths()
except (ValueError, FileNotFoundError) as e:
log.error(str(e))
return
output_path = self._generate_cache_folder(metadata, file)
if not output_path:
raise ValueError("Failed to generate cache folder path")
def run_gaia():
if os.path.exists(os.path.join(output_path, "gaia.json")):
log.debug(f"Gaia output already exists at {os.path.join(output_path, 'gaia.json')}, skipping")
return
subprocess.run(
[gaia_binary_path, file, os.path.join(output_path, "gaia.json")],
capture_output=True,
text=True,
env=ENV
)
gaia_thread = threading.Thread(target=run_gaia)
gaia_thread.start()
self._run_musicnn_models(REQUIRED_MODELS, musicnn_binary_path, file, output_path)
gaia_thread.join()
def analyze_optional(self, metadata: dict, file: str) -> None:
if not self._check_binaries():
log.error("Essentia binaries not found")
return
if not self._check_optional_models():
log.error("Optional models not found")
return
try:
musicnn_binary_path, _ = self._get_binary_paths()
except (ValueError, FileNotFoundError) as e:
log.error(str(e))
return
output_path = self._generate_cache_folder(metadata, file)
if not output_path:
raise ValueError("Failed to generate cache folder path")
self._run_musicnn_models(OPTIONAL_MODELS, musicnn_binary_path, file, output_path)
def _generate_cache_folder(self, metadata: dict, file_path: str) -> str:
cache_base = config.setting["acousticbrainz_ng_cache_path"]
if not cache_base:
raise ValueError("Cache path not configured")
release_artist_mbid = metadata.get('musicbrainz_albumartistid', 'NO_MBID')
release_group_mbid = metadata.get('musicbrainz_releasegroupid', 'NO_MBID')
release_mbid = metadata.get('musicbrainz_albumid', 'NO_MBID')
recording_mbid = metadata.get('musicbrainz_recordingid')
if not recording_mbid:
recording_mbid = self._get_audio_hash(file_path)
cache_folder = os.path.join(
str(cache_base),
str(release_artist_mbid),
str(release_group_mbid),
str(release_mbid),
str(recording_mbid)
)
os.makedirs(cache_folder, exist_ok=True)
return cache_folder
def _get_audio_hash(self, file_path: str) -> str:
try:
binaries_path = config.setting["acousticbrainz_ng_binaries_path"]
if not binaries_path:
raise ValueError("Binaries path not configured")
binary_path = self._get_binary_path("streaming_md5", binaries_path)
result = subprocess.run(
[binary_path, file_path],
capture_output=True,
text=True,
env=ENV
)
if result.returncode == 0:
for line in result.stdout.strip().split('\n'):
if line.startswith('MD5:'):
return line.split('MD5:')[1].strip()
log.error(f"Failed to calculate audio hash: {result.stderr}")
except Exception as e:
log.error(f"Error calculating audio hash: {e}")
return f"fallback_{hashlib.md5(file_path.encode('utf-8')).hexdigest()}"
def _check_binaries(self) -> bool:
path = config.setting["acousticbrainz_ng_binaries_path"]
if not path or not os.path.exists(path):
return False
for binary in REQUIRED_BINARIES:
binary_path = self._get_binary_path(binary, path)
if not os.path.exists(binary_path):
return False
return True
def _check_models(self, models: list[tuple[str, str]]) -> bool:
path = config.setting["acousticbrainz_ng_models_path"]
if not path or not os.path.exists(path):
return False
for model in models:
model_path = os.path.join(path, f"{model[0]}.pb")
if not os.path.exists(model_path):
return False
return True
def _check_required_models(self) -> bool:
return self._check_models(REQUIRED_MODELS)
def _check_optional_models(self) -> bool:
return self._check_models(OPTIONAL_MODELS)
acousticbrainz_ng = AcousticBrainzNG()
class AcousticBrainzNGTrackAction(BaseAction):
NAME = f"Analyze with {PLUGIN_NAME}"
def callback(self, objs):
tracks = list(filter(lambda o: isinstance(o, Track), objs))
for track in tracks:
acousticbrainz_ng.analyze_required(track.metadata, track.files[0].filename)
if config.setting["acousticbrainz_ng_analyze_optional"]:
acousticbrainz_ng.analyze_optional(track.metadata, track.files[0].filename)
register_options_page(AcousticBrainzNGOptionsPage)
register_track_action(AcousticBrainzNGTrackAction())

BIN
bin/streaming_extractor_music Executable file
View File

Binary file not shown.

BIN
bin/streaming_md5 Executable file
View File

Binary file not shown.

BIN
bin/streaming_musicnn_predict Executable file
View File

Binary file not shown.

87
models/danceability-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,87 @@
{
"name": "danceability",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/danceability/danceability-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by danceability",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"danceable",
"not_danceable"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"size": "306 full tracks, 124/182 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.93
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/danceability-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

67
models/gender-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,67 @@
{
"name": "gender",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/gender/gender-musicnn-msd-2.pb",
"version": "1",
"description": "classification of vocal music by gender (male/female)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": ["female", "male"],
"model_types": ["frozen_model"],
"dataset": {
"name": "In-house MTG collection",
"size": "3311 full tracks, 1508/1803 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.88
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [187, 96]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [1, 2],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [1, 2],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [1, 100],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [1, 200],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/gender-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

67
models/genre_electronic-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,67 @@
{
"name": "genre electronic",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/genre_electronic/genre_electronic-musicnn-msd-2.pb",
"version": "1",
"description": "classification of electronic music by subgenres",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": ["ambient", "drum and bass", "house", "techno", "trance"],
"model_types": ["frozen_model"],
"dataset": {
"name": "In-house MTG collection",
"size": "250 track excerpts, 50 per genre",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.95
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [187, 96]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [1, 5],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [1, 5],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [1, 100],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [1, 200],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/genre_electronic-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

77
models/genre_rosamerica-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,77 @@
{
"name": "genre Rosamerica",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/genre_rosamerica/genre_rosamerica-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by genre",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"classical",
"dance",
"hiphop",
"jazz",
"pop",
"rhythm and blues",
"rock",
"speech"
],
"model_types": ["frozen_model"],
"dataset": {
"name": "In-house MTG collection created by a musicologist",
"citation": "@phdthesis{i2009audio,\n title={Audio content processing for automatic music genre classification: descriptors, databases, and classifiers},\n author={i Termens, Enric Guaus},\n year={2009},\n school={PhD thesis, Universitat Pompeu Fabra, Barcelona, Spain}\n}",
"size": "400 tracks, 50 per genre",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.92
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [187, 96]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [1, 8],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [1, 8],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [1, 100],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [1, 200],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/genre_rosamerica-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

79
models/genre_tzanetakis-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,79 @@
{
"name": "genre GTZAN",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/genre_tzanetakis/genre_tzanetakis-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by genre",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"blues",
"classical",
"country",
"disco",
"hiphop",
"jazz",
"metal",
"pop",
"reggae",
"rock"
],
"model_types": ["frozen_model"],
"dataset": {
"name": "the GTZAN Genre Collection",
"citation": "@article{tzanetakis2002musical,\n title={Musical genre classification of audio signals},\n author={Tzanetakis, George and Cook, Perry},\n journal={IEEE Transactions on speech and audio processing},\n volume={10},\n number={5},\n pages={293--302},\n year={2002},\n publisher={IEEE}\n}",
"size": "1000 track excerpts, 100 per genre",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.83
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [187, 96]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [1, 10],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [1, 10],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [1, 100],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [1, 200],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/genre_tzanetakis-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

88
models/mood_acoustic-musicnn-mtt-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood acoustic",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_acoustic/mood_acoustic-musicnn-mtt-2.pb",
"version": "1",
"description": "classification of music by type of sound (acoustic/non-acoustic)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"acoustic",
"non_acoustic"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "321 full tracks + excerpts, 193/128 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.93
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_acoustic-musicnn-mtt-2.pb Normal file
View File

Binary file not shown.

88
models/mood_aggressive-musicnn-mtt-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood aggressive",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_aggressive/mood_aggressive-musicnn-mtt-2.pb",
"version": "1",
"description": "classification of music by mood (aggressive/non-aggressive)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"aggressive",
"not_aggressive"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection ",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "280 full tracks + excerpts, 133/147 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.96
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_aggressive-musicnn-mtt-2.pb Normal file
View File

Binary file not shown.

88
models/mood_electronic-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood electronic",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_electronic/mood_electronic-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by type of sound (electronic/non-electronic)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"electronic",
"non_electronic"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "332 full tracks + excerpts, 164/168 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.95
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_electronic-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

88
models/mood_happy-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood happy",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_happy/mood_happy-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by mood (happy/non-happy)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"happy",
"non_happy"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "302 full tracks + excerpts, 139/163 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.81
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_happy-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

88
models/mood_party-musicnn-mtt-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood party",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_party/mood_party-musicnn-mtt-2.pb",
"version": "1",
"description": "classification of music by mood (party/non-party)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"non_party",
"party"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "349 full tracks + excerpts, 198/151 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.92
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_party-musicnn-mtt-2.pb Normal file
View File

Binary file not shown.

88
models/mood_relaxed-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood relaxed",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_relaxed/mood_relaxed-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by mood (relaxed/non-relaxed)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"non_relaxed",
"relaxed"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "446 full tracks + excerpts, 145/301 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.9
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_relaxed-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

88
models/mood_sad-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,88 @@
{
"name": "mood sad",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/mood_sad/mood_sad-musicnn-msd-2.pb",
"version": "1",
"description": "classification of music by mood (sad/non-sad)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"non_sad",
"sad"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"citation": "@inproceedings{laurier2009music,\n title={Music mood annotator design and integration},\n author={Laurier, Cyril and Meyers, Owen and Serra, Joan and Blech, Martin and Herrera, Perfecto},\n booktitle={2009 Seventh International Workshop on Content-Based Multimedia Indexing},\n pages={156--161},\n year={2009},\n organization={IEEE}\n}",
"size": "230 full tracks + excerpts, 96/134 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.86
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/mood_sad-musicnn-msd-2.pb Normal file
View File

Binary file not shown.

126
models/msd-musicnn-1.json Normal file
View File

@@ -0,0 +1,126 @@
{
"name": "MSD MusiCNN",
"type": "auto-tagging",
"link": "https://essentia.upf.edu/models/feature-extractors/musicnn/msd-musicnn-1.pb",
"version": "1",
"description": "prediction of the top-50 tags in the dataset",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-03-31",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"rock",
"pop",
"alternative",
"indie",
"electronic",
"female vocalists",
"dance",
"00s",
"alternative rock",
"jazz",
"beautiful",
"metal",
"chillout",
"male vocalists",
"classic rock",
"soul",
"indie rock",
"Mellow",
"electronica",
"80s",
"folk",
"90s",
"chill",
"instrumental",
"punk",
"oldies",
"blues",
"hard rock",
"ambient",
"acoustic",
"experimental",
"female vocalist",
"guitar",
"Hip-Hop",
"70s",
"party",
"country",
"easy listening",
"sexy",
"catchy",
"funk",
"electro",
"heavy metal",
"Progressive rock",
"60s",
"rnb",
"indie pop",
"sad",
"House",
"happy"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "The Millon Song Dataset",
"citation": "http://millionsongdataset.com/",
"size": "200k up to two minutes audio previews",
"metrics": {
"ROC-AUC": 0.88,
"PR-AUC": 0.29
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
50
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
50
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/msd-musicnn-1.pb Normal file
View File

Binary file not shown.

87
models/tonal_atonal-musicnn-mtt-2.json Normal file
View File

@@ -0,0 +1,87 @@
{
"name": "tonal/atonal",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/tonal_atonal/tonal_atonal-musicnn-mtt-2.pb",
"version": "1",
"description": "classification of music by tonality (tonal/atonal)",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"atonal",
"tonal"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"size": "345 track excerpts, 145/200 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.91
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/tonal_atonal-musicnn-mtt-2.pb Normal file
View File

Binary file not shown.

87
models/voice_instrumental-musicnn-msd-2.json Normal file
View File

@@ -0,0 +1,87 @@
{
"name": "voice/instrumental classifiers",
"type": "multi-class classifier",
"link": "https://essentia.upf.edu/models/classifiers/voice_instrumental/voice_instrumental-musicnn-msd-2.pb",
"version": "1",
"description": "classification into music with voice/instrumental",
"author": "Pablo Alonso",
"email": "pablo.alonso@upf.edu",
"release_date": "2020-07-07",
"framework": "tensorflow",
"framework_version": "1.15.0",
"classes": [
"instrumental",
"voice"
],
"model_types": [
"frozen_model"
],
"dataset": {
"name": "In-house MTG collection",
"size": "1000 track excerpts, 500 per class",
"metrics": {
"5-fold_cross_validation_normalized_accuracy": 0.98
}
},
"schema": {
"inputs": [
{
"name": "model/Placeholder",
"type": "float",
"shape": [
187,
96
]
}
],
"outputs": [
{
"name": "model/Sigmoid",
"type": "float",
"shape": [
1,
2
],
"op": "Sigmoid",
"output_purpose": "predictions"
},
{
"name": "model/dense_2/BiasAdd",
"type": "float",
"shape": [
1,
2
],
"op": "fully connected",
"description": "logits",
"output_purpose": ""
},
{
"name": "model/dense_1/BiasAdd",
"type": "float",
"shape": [
1,
100
],
"op": "fully connected",
"description": "penultimate layer",
"output_purpose": ""
},
{
"name": "model/dense/BiasAdd",
"type": "float",
"shape": [
1,
200
],
"op": "fully connected",
"output_purpose": "embeddings"
}
]
},
"citation": "@inproceedings{alonso2020tensorflow,\n title={Tensorflow Audio Models in Essentia},\n author={Alonso-Jim{\\'e}nez, Pablo and Bogdanov, Dmitry and Pons, Jordi and Serra, Xavier},\n booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n year={2020}\n}",
"inference": {
"sample_rate": 16000,
"algorithm": "TensorflowPredictMusiCNN"
}
}

BIN
models/voice_instrumental-musicnn-msd-2.pb Normal file
View File

Binary file not shown.