This repository contains the training and evaluation code of our state-of-the-art version identification system along with pre-trained models. For a detailed explanation of our design decisions and the general pipeline, please refer to our publication.
Furkan Yesiler, Joan Serrà and Emilia Gómez, "Accurate and scalable version identification using musically-motivated embeddings," in Proc. of the IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP), 2020.
Below, we specify some use cases and explain the important steps you need to follow for using the code in this repository.
To facilitate the benchmarking process and to present a pipeline for evaluating the pre-trained MOVE models, we have prepared benchmark_da-tacos.py. To use the script, you can follow the steps below:
git clone https://github.com/furkanyesiler/move.git
cd move
python3 -m venv venv
source venv/bin/activate
pip install -r requirements_benchmark.txt
After creating the virtual environment and installing the required packages, you can simply run
python benchmark_da-tacos.py --unpack --remove
usage: benchmark_da-tacos.py [-h] [--outputdir OUTPUTDIR] [--unpack]
[--remove]
Downloading and preprocessing cremaPCP features of the Da-TACOS benchmark
subset
optional arguments:
-h, --help show this help message and exit
--outputdir OUTPUTDIR
Directory to store the dataset (default: ./data)
--unpack Unpack the zip files (default: False)
--remove Remove zip files after unpacking (default: False)
This script downloads the metadata and the cremaPCP features of the Da-TACOS benchmark subset, and preprocesses them to work with our evaluation setting. Specifically, after downloading the files:
.pt
file,Both the data and the ground truth annotations (named benchmark_crema.pt
and ytrue_benchmark.pt
, respectively) are stored in the data
folder.
For this use case, we would like to point out a number of requirements you must follow.
git clone https://github.com/furkanyesiler/move.git
cd move
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
MOVE was trained using cremaPCP features, and therefore, it may underperform with other Pitch Class Profile (PCP, or chroma) variants. To extract cremaPCP features for your audio collection, please refer to acoss.
cremaPCP features used for training are created using non-overlapping frames and a hop size of 4096 on audio tracks sampled at 44.1 kHz (about 93 ms per frame).
After obtaining cremaPCP features for your dataset, you should cast them as a torch.Tensor, and reshape them from Tx12 to 1x23xT. For this step, you can use the following code snippet:
import numpy as np
import torch
# this variable represents your cremaPCP feature
cremaPCP = np.random.rand(100, 12)
# this is casting the crema feature to a torch.Tensor type
cremaPCP_tensor = torch.from_numpy(cremaPCP).t()
# this is the resulting cremaPCP feature
cremaPCP_reshaped = torch.cat((cremaPCP_tensor, cremaPCP_tensor))[:23].unsqueeze(0)
After the features are downloaded and preprocessed, you can use the script below to evaluate the pre-trained MOVE model on the Da-TACOS benchmark subset:
python move_main.py -rt test --dataset 1
or
python move_main.py -rt test --dataset 1 -emb 4000
When the cremaPCP features per song are ready, you need to create a dataset file and a ground truth annotations file. The dataset file should be a python dictionary with 2 keys: data
and labels
. Each key (i.e. 'data' or 'labels') should point to a python list which contains the respective cremaPCP features and label of that track. Specifically, let dataset_dict
be our dataset dictionary, and dataset_dict['data']
and dataset_dict['labels']
be our lists. The label of the song dataset_dict['data'][42]
should be dataset_dict['labels'][42]
. Finally, the dataset file should be saved under data
folder, and should be named benchmark_crema.pt
. An example code is shown below:
import os
root_dir = '/your/root/directory/of/move'
data = []
labels = []
for i in range(dataset_size):
cremaPCP = load_cremaPCP(i) # loading the cremaPCP features for the ith song of your dataset
label = load_label(i) # loading the label of the ith song of your dataset
data.append(cremaPCP)
labels.append(label)
dataset_dict = {'data': data, 'labels': label}
torch.save(dataset_dict, os.path.join(root_dir, 'data', 'benchmark_crema.pt'))
When your dataset file ('benchmark_crema.pt') is ready, you have to create a ground truth annotations file which is stored in data
folder, and should be named ytrue_benchmark.pt
. This file should be a torch.Tensor with the shape NxN (N is the size of your dataset). Finally, the diagonal of this matrix should be 0. You can find an example code below:
import os
import torch
data_dir = '/your/root/directory/of/move/data/'
labels = torch.load(os.path.join(data_dir, 'benchmark_crema.pt'))['labels']
ytrue = []
for i in range(len(labels)):
main_label = labels[i] # label of the ith song
sub_ytrue = []
for j in range(len(labels)):
if labels[j] == main_label and i!= j: # checking whether the ith and jth song has the same label
sub_ytrue.append(1)
else:
sub_ytrue.append(0)
ytrue.append(sub_ytrue)
ytrue = torch.Tensor(ytrue)
torch.save(ytrue, os.path.join(data_dir, 'ytrue_benchmark.pt'))
After you prepared your dataset and annotation files, you can use the script below to evaluate the pre-trained MOVE model on your dataset:
python move_main.py -rt test --dataset 1
or
python move_main.py -rt test --dataset 1 -emb 4000
For training MOVE with a private dataset, you should follow the steps below:
git clone https://github.com/furkanyesiler/move.git
cd move
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
The main requirement for the training is to create dataset file(s) and an annotation file for the validation set. For this, you should extract cremaPCP features for the songs in your dataset, and apply the preprocessing steps described at 2.2.
The dataset files should be created with the same structure as explained in 2.2.1. They should be python dictionaries with two keys (i.e. 'data'
and 'labels'
), and should be placed under data
folder.
If your training data includes more than one file, then you should use the following naming convention:
If your dataset is named 'my_training_set', and is divided into 3 files, you should name them:
my_training_set_1.pt
my_training_set_2.pt
my_training_set_3.pt
In this case, when you run move_main.py, you should use the following arguments:
python move_main.py --train_path my_training_set --chunks 3
Current version of our code does not support using more than one file for the validation set.
The training procedure of MOVE tracks Mean Average Precision (MAP) score on the validation set for each epoch. To compute MAP, you need to provide an annotations file under the data
folder. You can see an example of how to create such file at 2.2.2. The name of your annotations file for your validation set should be ytrue_validation.pt
.
After preparing the dataset and annotations files, you can run the training procedure with the following command:
python move_main.py
usage: move_main.py [-h] [-rt {train,test}] [-tp TRAIN_PATH] [-ch CHUNKS]
[-vp VAL_PATH] [-sm {0,1}] [-ss {0,1}] [-rs RANDOM_SEED]
[-noe NUM_OF_EPOCHS] [-m {0,1}] [-emb EMB_SIZE]
[-sum {0,1,2,3,4}] [-fa {0,1,2,3}] [-lr LEARNING_RATE]
[-lrs {0,1,2}] [-lrsf LRSCH_FACTOR] [-mo MOMENTUM]
[-pl PATCH_LEN] [-nol NUM_OF_LABELS] [-da {0,1}]
[-nd {0,1}] [-ms {0,1,2}] [-ma MARGIN] [-ytc {0,1}]
[-d {0,1,2}] [-dn DATASET_NAME]
Training code of MOVE
optional arguments:
-h, --help show this help message and exit
-rt {train,test}, --run_type {train,test}
Whether to run train or test script
-tp TRAIN_PATH, --train_path TRAIN_PATH
Path for training data. If more than one file are
used, write only the common part
-ch CHUNKS, --chunks CHUNKS
Number of chunks for training set
-vp VAL_PATH, --val_path VAL_PATH
Path for validation data
-sm {0,1}, --save_model {0,1}
1 for saving the trained model, 0 for otherwise
-ss {0,1}, --save_summary {0,1}
1 for saving the training log, 0 for otherwise
-rs RANDOM_SEED, --random_seed RANDOM_SEED
Random seed
-noe NUM_OF_EPOCHS, --num_of_epochs NUM_OF_EPOCHS
Number of epochs for training
-m {0,1}, --model_type {0,1}
0 for MOVE, 1 for MOVE without pitch transposition
-emb EMB_SIZE, --emb_size EMB_SIZE
Size of the final embeddings
-sum {0,1,2,3,4}, --sum_method {0,1,2,3,4}
0 for max-pool, 1 for mean-pool, 2 for autopool, 3 for
multi-channel attention, 4 for multi-channel adaptive
attention
-fa {0,1,2,3}, --final_activation {0,1,2,3}
0 for no activation, 1 for sigmoid, 2 for tanh, 3 for
batch norm
-lr LEARNING_RATE, --learning_rate LEARNING_RATE
Initial learning rate
-lrs {0,1,2}, --lr_schedule {0,1,2}
0 for no lr_schedule, 1 for decreasing lr at epoch 80,
2 for decreasing lr at epochs [80, 100]
-lrsf LRSCH_FACTOR, --lrsch_factor LRSCH_FACTOR
Factor for lr scheduler
-mo MOMENTUM, --momentum MOMENTUM
Value for momentum parameter for SGD
-pl PATCH_LEN, --patch_len PATCH_LEN
Size of the input len in time dimension
-nol NUM_OF_LABELS, --num_of_labels NUM_OF_LABELS
Number of cliques per batch for triplet mining
-da {0,1}, --data_aug {0,1}
0 for no data aug, 1 using it
-nd {0,1}, --norm_dist {0,1}
1 for normalizing the distance, 0 for avoiding it
-ms {0,1,2}, --mining_strategy {0,1,2}
0 for only random, 1 for only semi-hard, 2 for only
hard
-ma MARGIN, --margin MARGIN
Margin for triplet loss
-ytc {0,1}, --ytc_labels {0,1}
0 for using full training data, 1 for removing
overlapping labels with ytc
-d {0,1,2}, --dataset {0,1,2}
Choosing evaluation set for testing. 0 for move
validation, 1 for test on da-tacos, 2 for test on ytc
-dn DATASET_NAME, --dataset_name DATASET_NAME
Specifying a dataset name for evaluation. The dataset
must be located in the data folder
For any questions you may have, feel free to create an issue or contact me.
The code in this repository is licensed under Affero GPL v3.
Please cite our reference if you plan to use the code in this repository:
@inproceedings{yesiler2020,
author = "Furkan Yesiler and Joan Serrà and Emilia Gómez",
title = "Accurate and scalable version identification using musically-motivated embeddings",
booktitle = "Proc. of the IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP)",
year = "2020"
}
This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 765068 (MIP-Frontiers).
This work has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No. 770376 (TROMPA).