Skip to content
Snippets Groups Projects
Commit b56a8e66 authored by MARMORET Axel's avatar MARMORET Axel
Browse files

Modifying the 'features' into 'signal_to_spectrogram', following as-seg

parent ba66c218
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
barmuscomp/model/features.py → barmuscomp/model/spectrogram_to_signal.py
+ 147
− 147
View file @ b56a8e66
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 25 16:54:59 2020
@author: amarmore
"""
import barmuscomp.model.errors as err
# %% Load everything from as_seg
from as_seg.model.features import *
# See details in as_seg
# %% HCQT
def get_hcqt_params():
"""
Credit to & al. [1] (comes directly from https://github.com/rabitt/ismir2017-deepsalience)
Fixing parameters for the HCQT computation.
Returns
-------
bins_per_octave : TYPE
DESCRIPTION.
n_octaves : TYPE
DESCRIPTION.
harmonics : TYPE
DESCRIPTION.
sr : TYPE
DESCRIPTION.
fmin : TYPE
DESCRIPTION.
hop_length : TYPE
DESCRIPTION.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
bins_per_octave = 60
n_octaves = 6
harmonics = [0.5, 1, 2, 3, 4, 5]
sr = 22050
fmin = 32.7
hop_length = 256
return bins_per_octave, n_octaves, harmonics, sr, fmin, hop_length
def compute_hcqt_bittner(signal, sr):
"""
Credit to Bittner & al. [1] (comes from https://github.com/rabitt/ismir2017-deepsalience).
Computes HCQT representation of the signal, as presented in [1] (3-rd order tensor).
Parameters
----------
signal : numpy array
Signal of the song.
sr : int
the sampling_rate
Returns
-------
log_hcqt : np array
The tensor of logarithm HCQT.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
(bins_per_octave, n_octaves, harmonics,
sr, f_min, hop_length) = get_hcqt_params()
#y, fs = librosa.load(audio_fpath, sr=sr)
cqt_list = []
shapes = []
for h in harmonics:
cqt = librosa.cqt(
signal, sr=sr, hop_length=hop_length, fmin=f_min*float(h),
n_bins=bins_per_octave*n_octaves,
bins_per_octave=bins_per_octave
)
cqt_list.append(cqt)
shapes.append(cqt.shape)
shapes_equal = [s == shapes[0] for s in shapes]
if not all(shapes_equal):
min_time = np.min([s[1] for s in shapes])
new_cqt_list = []
for i in range(len(cqt_list)):
new_cqt_list.append(cqt_list[i][:, :min_time])
cqt_list = new_cqt_list
log_hcqt = ((1.0/80.0) * librosa.core.amplitude_to_db(
np.abs(np.array(cqt_list)), ref=np.max)) + 1.0
return log_hcqt
def my_compute_hcqt(signal, sr):
"""
Credit to Bittner & al. [1] (comes from https://github.com/rabitt/ismir2017-deepsalience).
Computes HCQT representation of the signal, as presented in [1] (3-rd order tensor).
The order of the mode is changed though, so tht first two modes correspond to frequency and time respectively,
and that the third corresponds to harmonic content.
Parameters
----------
signal : numpy array
Signal of the song.
sr : int
the sampling_rate
Returns
-------
log_hcqt : np array
The tensor of logarithm HCQT.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
(bins_per_octave, n_octaves, harmonics, sr, f_min, hop_length) = get_hcqt_params()
freq_mode_len = bins_per_octave*n_octaves
first_cqt = librosa.cqt(signal, sr=sr, hop_length=hop_length, fmin=f_min*float(harmonics[0]),
n_bins=freq_mode_len, bins_per_octave=bins_per_octave)
time_mode_len = first_cqt.shape[1]
h_cqt = np.array(first_cqt).reshape(freq_mode_len, time_mode_len, 1)
for h in harmonics[1:]:
cqt = librosa.cqt(signal, sr=sr, hop_length=hop_length, fmin=f_min*float(h),
n_bins=bins_per_octave*n_octaves,bins_per_octave=bins_per_octave)
current_cqt = cqt.reshape(freq_mode_len, time_mode_len, 1)
h_cqt = np.append(h_cqt, current_cqt, axis = 2)
log_hcqt = ((1.0/80.0) * librosa.core.amplitude_to_db(np.abs(h_cqt), ref=np.max)) + 1.0
return log_hcqt
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 25 16:54:59 2020
@author: amarmore
"""
import barmuscomp.model.errors as err
# %% Load everything from as_seg
from as_seg.model.spectrogram_to_signal import *
# See details in as_seg
# %% HCQT
def get_hcqt_params():
"""
Credit to & al. [1] (comes directly from https://github.com/rabitt/ismir2017-deepsalience)
Fixing parameters for the HCQT computation.
Returns
-------
bins_per_octave : TYPE
DESCRIPTION.
n_octaves : TYPE
DESCRIPTION.
harmonics : TYPE
DESCRIPTION.
sr : TYPE
DESCRIPTION.
fmin : TYPE
DESCRIPTION.
hop_length : TYPE
DESCRIPTION.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
bins_per_octave = 60
n_octaves = 6
harmonics = [0.5, 1, 2, 3, 4, 5]
sr = 22050
fmin = 32.7
hop_length = 256
return bins_per_octave, n_octaves, harmonics, sr, fmin, hop_length
def compute_hcqt_bittner(signal, sr):
"""
Credit to Bittner & al. [1] (comes from https://github.com/rabitt/ismir2017-deepsalience).
Computes HCQT representation of the signal, as presented in [1] (3-rd order tensor).
Parameters
----------
signal : numpy array
Signal of the song.
sr : int
the sampling_rate
Returns
-------
log_hcqt : np array
The tensor of logarithm HCQT.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
(bins_per_octave, n_octaves, harmonics,
sr, f_min, hop_length) = get_hcqt_params()
#y, fs = librosa.load(audio_fpath, sr=sr)
cqt_list = []
shapes = []
for h in harmonics:
cqt = librosa.cqt(
signal, sr=sr, hop_length=hop_length, fmin=f_min*float(h),
n_bins=bins_per_octave*n_octaves,
bins_per_octave=bins_per_octave
)
cqt_list.append(cqt)
shapes.append(cqt.shape)
shapes_equal = [s == shapes[0] for s in shapes]
if not all(shapes_equal):
min_time = np.min([s[1] for s in shapes])
new_cqt_list = []
for i in range(len(cqt_list)):
new_cqt_list.append(cqt_list[i][:, :min_time])
cqt_list = new_cqt_list
log_hcqt = ((1.0/80.0) * librosa.core.amplitude_to_db(
np.abs(np.array(cqt_list)), ref=np.max)) + 1.0
return log_hcqt
def my_compute_hcqt(signal, sr):
"""
Credit to Bittner & al. [1] (comes from https://github.com/rabitt/ismir2017-deepsalience).
Computes HCQT representation of the signal, as presented in [1] (3-rd order tensor).
The order of the mode is changed though, so tht first two modes correspond to frequency and time respectively,
and that the third corresponds to harmonic content.
Parameters
----------
signal : numpy array
Signal of the song.
sr : int
the sampling_rate
Returns
-------
log_hcqt : np array
The tensor of logarithm HCQT.
References
----------
[1] Bittner, R. M., McFee, B., Salamon, J., Li, P., & Bello, J. P. (2017, October).
Deep Salience Representations for F0 Estimation in Polyphonic Music. In ISMIR (pp. 63-70).
"""
(bins_per_octave, n_octaves, harmonics, sr, f_min, hop_length) = get_hcqt_params()
freq_mode_len = bins_per_octave*n_octaves
first_cqt = librosa.cqt(signal, sr=sr, hop_length=hop_length, fmin=f_min*float(harmonics[0]),
n_bins=freq_mode_len, bins_per_octave=bins_per_octave)
time_mode_len = first_cqt.shape[1]
h_cqt = np.array(first_cqt).reshape(freq_mode_len, time_mode_len, 1)
for h in harmonics[1:]:
cqt = librosa.cqt(signal, sr=sr, hop_length=hop_length, fmin=f_min*float(h),
n_bins=bins_per_octave*n_octaves,bins_per_octave=bins_per_octave)
current_cqt = cqt.reshape(freq_mode_len, time_mode_len, 1)
h_cqt = np.append(h_cqt, current_cqt, axis = 2)
log_hcqt = ((1.0/80.0) * librosa.core.amplitude_to_db(np.abs(h_cqt), ref=np.max)) + 1.0
return log_hcqt
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment