3.02 Sync voices¶
- The voices are now synchronized
- Each voice has an own rolling mean window and scale factor
- Functions which are no longer part of this development step are exported to the music_generation.py file. The file is found at the end of the page.
from pyknon.genmidi import Midi
from pyknon.music import Rest, Note, NoteSeq
from music_generation import *
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.ticker as plticker
from datetime import date
Transform Meteo data¶
def scale(a): return (a-a.min())/(a.max()-a.min())
def read_meteo_data(fName):
colNames = ['Stao','time', 'Flash', 'p_QNH', 'T_2m', 'Precip', 'H_rel', 'V_wind']
df = pd.read_csv(fName,sep=';', skiprows=3, names=colNames, na_values='-')
print(df.head())
return df
fPath = '/mnt/daten/04_Schule/42_Kanti/Matrua/Music_generation/Organisation/MeteoSchweiz/Daten/'
fName = 'order_74678_data.txt'
dM = read_meteo_data(fPath+fName)
#---- Parameter bestimmen -----------
NT, MP = dM.shape
print('-----------------')
print('NT, MP', NT, MP)
Stao time Flash p_QNH T_2m Precip H_rel V_wind
0 KLO 201908280000 0 968.5 19.6 0.0 90.1 0.6
1 KLO 201908280010 0 968.5 19.3 0.0 93.0 0.6
2 KLO 201908280020 0 968.6 19.4 0.0 90.6 0.7
3 KLO 201908280030 0 968.7 19.6 0.0 90.3 0.7
4 KLO 201908280040 0 968.7 18.7 0.0 95.6 0.5
-----------------
NT, MP 2016 8
met_transform
- the rolling mean is to remove noise on the data.
- the factors are used to scale the melody, such that it plays in a certain range
- start defines the staring point of the melodies by removing the begin of the data
def met_transform(dM,factors,means,start):
col_nr = dM.shape[1]-2
start = int(start*6)
cut_border = np.trunc((np.amax(means))/2).astype(int) # calculate nr of nan at the border because of the rolling mean
cut_begin = np.amax([cut_border,start])
trans = np.zeros((col_nr, (dM.shape[0] -cut_border -cut_begin)))
if col_nr != len(factors) or col_nr != len(means): print('dM,factor,mean not same length')
for nr,factor, mean in zip(range(col_nr),factors,means):
Yw = np.array(dM[dM.columns[nr +2]].rolling(window=mean,center=True).mean()) # nr+2 the first two colums are location and date.
Yw = Yw * factor
trans[nr] = Yw[cut_begin: -cut_border] # remove nan at begining and end. because of rolling mean
return trans
Chords and scales
major = np.array([ 0, 2, 4, 5, 7, 9, 11])
minor = np.array([ 0, 2, 3, 5, 7, 8, 10])
C7 = np.array([ 0, 4, 7, 10])
CM7 = np.array([ 0, 4, 7, 11])
Cm7 = np.array([ 0, 3, 7, 10])
Cm = np.array([ 0, 3, 7])
C = np.array([ 0, 4, 7])
bass= np.array([ 0])
Tune U¶
This tune uses the wind and temperature data, starting after 80 hours –> 30.8.2019
def meteo_melody(meteo, pattern, start_note, a_range, notenr, rythem,mpb):
melody = np.zeros(notenr, dtype=int)
cum_rythem = np.cumsum(rythem) *4
cum_rythem = np.concatenate(([0],cum_rythem)) # add 0 at beginig
scale_change = pattern[:,0]
scale_nr =0
scale = pattern[scale_nr,1]
melody[0] = scale[i_last_note(start_note,scale)]
for npn in range(1, notenr): #npn: note per note (index)
scale_nr = np.ravel(np.argwhere(scale_change <= cum_rythem[npn-1])) [-1]
scale = pattern[scale_nr,1]
# find interval
met_resolution = 10
inter = np.asarray([cum_rythem[npn-1], cum_rythem[npn]]) # get beat_nr's
inter = np.round((inter*mpb)/met_resolution).astype(int) # calulate index of the data array
intvl = meteo[inter[1]] - meteo[inter[0]] # take the diffrence of the data
intvl = np.round(intvl).astype(int) # round to an int
inote_befor = i_last_note(melody[npn-1],scale) # get i in the scale of the last note
inote = inote_befor + intvl # calculate i in scale of note
melody[npn] = scale[inote] # set in to melody
#print(melody)
plt.plot(cum_rythem[1:],melody) ; plt.xlabel= ('beat nr.'); plt.ylabel=('midi note nr')
return melody
def tune_U():
tune_name = 'tune_U'
#np.random.seed(23)
bar, bpb = 12, 4 # bar: Takt , bpb: beat per bar
melody_len = bar * bpb
mpb = 70 #minutes per beat.
start = 79.5 # start in hours
trans = met_transform(dM,[1,2.5,0.8,1,0.3,4.5],[6,6,6,6,6,2],start)
#plt.plot(trans[5,:300])
#np.set_printoptions(threshold=np.inf)
#print(trans[1,::20])
scales = [[1,CM7],[1,Cm7+9],[1,Cm7+2],[1,C7+7]] #rythem Change
#scales = [[4,C7],[2,C7+5],[2,C7],[1,C7+7],[1,C7+5],[2,C7]] # Blues
scales = [[8,minor]]
pattern = pattern_gen(scales, melody_len)
range_1 = liniar_range(44,51,70,76)
rythem1, notenr_1 = ran_duration([1/16,1/8, 1/4,1/2], [2,4,1,0], melody_len)
melody1 = meteo_melody(trans[5],pattern, 72, range_1, notenr_1, rythem1,mpb)
volumes1 = ran_volume([0,120], [1,8], notenr_1 )
notes1 = NoteSeq( [Note(no,octave=0, dur=du, volume=vo) for no,du,vo in zip(melody1,rythem1,volumes1)] )
range_2 = liniar_range(44,51,70,76)
rythem2, notenr_2 = ran_duration([1/16,1/8, 1/4,1/2], [0,2,3,2], melody_len)
melody2 = meteo_melody(trans[4],pattern, 65, range_2, notenr_2, rythem2,mpb)
volumes2 = ran_volume([0,120], [1,8], notenr_2 )
notes2 = NoteSeq( [Note(no,octave=0, dur=du, volume=vo) for no,du,vo in zip(melody2,rythem2,volumes2)] )
#plot_range([range_1],['range_1'],tune_name)
instruments = [10,49]
notes = [notes1,notes2]
return notes, instruments,tune_name
tune_U
tune_U tune_U_2
tune_U_2 Instruments: Available are at lest the 128 General-Midi (GM) Instruments. Depending on the sound-fonts there is a bigger choice. A list of the GM instruments can be found here. https://jazz-soft.net/demo/GeneralMidi.html
Generate Midi and Audio file¶
def gen_midi():
# squezze into a MIDI framework
notes, instruments, tune_name = tune_U() # <--- select a tune <<-- <<<<<<<<<--- select a tune -----
nTracks = len(notes)
m = Midi(number_tracks=nTracks, tempo=120, instrument=instruments)
for iTrack in range(nTracks):
m.seq_notes(notes[iTrack], track=iTrack)
#--- write the MIDI file -----
midi_file_name = tune_name +'.mid' # set the name of the file
m.write(midi_file_name)
return midi_file_name
######--- Main ---######
midi_file_name = gen_midi()
midi_play(midi_file_name)
midi_audio(midi_file_name)
midi_png(midi_file_name)
External Music_Generation library¶
This library changes from version to version. New or changed code is first explained above. This is a copy of music_generation.py
from pyknon.genmidi import Midi
from pyknon.music import Rest, Note, NoteSeq
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.ticker as plticker
from datetime import date
# [[[[[[[[[[[[[[[[[[[ -- Functions for Music Generation -- ]]]]]]]]]]]]]]]]]]]
def scale_create(tones):
tones = np.asarray(tones) # tones which form chord or scale in the first octave (0-11)
if any(tones > 11): # tones over one octave?
tones = np.mod(tones,12) # set the thones in one octave
tones = np.sort(tones) # sort the tones new
tones = np.unique(tones) # remove duplicate tones
octave = np.repeat( np.linspace(0,108, num=10), len(tones))
scale = np.add( octave, np.tile(tones, 10)) # add element wise octave and note
return scale.astype(int)
def fade(start,end,steps):
fade = np.around( np.linspace(start,end,num=steps))
fade = fade.astype(int)
return fade
def ran_volume(volume, prob_volume, melody_len):
volume = np.asarray(volume, dtype=int) # this are the allowed volumes of thenotes
prob_volume = np.asarray(prob_volume) # this are the probabilities how often each volume will occure
prob_volume = prob_volume/np.sum(prob_volume)
volumes = np.r_[np.random.choice(volume, size=melody_len, p=prob_volume)]
return volumes
# liniar_range: Generates an range in which the instrument can play.
def liniar_range(r_start, r_top, r_edge, r_end): # acceptance range of the instrument
h = 100 # hight of acceptance function
a_range = np.zeros(121, dtype=int) # only to midi =120 as 127 is not a complete octave
np.put(a_range, range(r_start,r_top), np.linspace(0,h, num=(r_top -r_start)) )
np.put(a_range, range(r_top, r_edge), np.linspace(h,h, num=(r_edge-r_top )) )
np.put(a_range, range(r_edge, r_end), np.linspace(h,0, num=(r_end -r_edge )) )
return a_range
# i_last_note: finds de i value of the last not in the actual scale.
def i_last_note(note, scale):
i_note = (np.abs(scale - note)).argmin()
return i_note
# intvl_next is a modification of intvl_melody. But it does only creats one interval and not an array/melody in one time.
def intvl_next(intvl, prob_intvl): #singel interval
intvl = np.asarray(intvl) # Possible interval
prob_intvl = np.asarray(prob_intvl) # Probability of each interval
prob_intvl = prob_intvl/np.sum(prob_intvl)
interval = np.random.choice(intvl, size=1, p=prob_intvl)
return interval[0]
# acceptance: accepts and refuses proposed nots with Metropolis-Hasting Algorythem.
# x is the value in the aceptance range of the current note, while x_new is it from the proposoal note
def acceptance(x, x_new):
if x_new < 1:
if x < 1: print('start_note not in range') ; x = start_note_not_in_range
quot = x_new/x
if quot >= 1: return True
if np.random.uniform(0,1)< quot: return True
else: return False
def ran_duration(duration, prob_duration, melody_len):
duration= np.asarray(duration) # this are the allowed durations of the notes
prob_duration = np.asarray(prob_duration) # this are the probabilities how often each will occure
prob_duration = prob_duration/np.sum(prob_duration)
cumsum, melody_len, rythem = 0, melody_len/4 , np.asarray([]) #melody_len/4 as note values are quarter
while cumsum < melody_len:
note_len = np.random.choice(duration, p=prob_duration)
cumsum = cumsum + note_len
rythem = np.append(rythem,note_len)
return rythem , len(rythem)
# pattern_gen takes the chord pattern (scales): it reapeats the pattern as long the melody is, and generates the beat number where the chords change.
def pattern_gen(scales,melody_len):
scales = np.asarray(scales)
bpb = 4 # beats per bar
factor = int(np.trunc(melody_len/(np.sum(scales[:,0]) * bpb)) + 1) # factor rounded up: how many times is the pattern used
change_times = np.cumsum(np.tile(scales[:,0],factor)) * bpb # create change time list with factor
change_times = np.concatenate((np.asarray([0]),change_times))[:-1] # add 0 at beginig remove last element
for i in range(len(scales)): # send scales to scale_create
scales[i,1] = scale_create(scales[i,1])
pattern = np.tile(scales,(factor,1)) # tile the scales as long the melody is
pattern[:,0] = change_times #insert change_times into scales
pattern = np.delete(pattern, np.argwhere(pattern[:,0] >= melody_len) ,0) # remove unneeded scales
return pattern
def acceptance_melody(intvl, prob_intvl, pattern, start_note, a_range, notenr, rythem):
melody = np.zeros(notenr, dtype=int)
cum_rythem = np.cumsum(rythem) *4
cum_rythem = np.concatenate(([0],cum_rythem))[:-1] # add 0 at beginig remove last element
scale_change = pattern[:,0]
scale_nr =0
scale = pattern[scale_nr,1]
melody[0] = scale[i_last_note(start_note,scale)]
for npn in range(1, notenr): #npn: note per note (index)
scale_nr = np.ravel(np.argwhere(scale_change <= cum_rythem[npn-1])) [-1]
scale = pattern[scale_nr,1]
accept = False
while not accept: # aslong acept == False
inote = i_last_note(melody[npn-1],scale)
inote_next = inote + intvl_next(intvl, prob_intvl) # add current not with Proposition
accept_val = a_range[[melody[(npn-1)],scale[inote_next]]] # get acceptance values
accept = acceptance(accept_val[0],accept_val[1])
melody[npn] = scale[inote_next]
return melody
# plot_range: plot all ranges together
def plot_range(ranges,labels,title):
fig, ax = plt.subplots()
plt.xlabel('Midi Note')
plt.ylabel('Acceptance')
plt.title(title)
for a_range, lab in zip(ranges,labels):
ax.plot(range(121), a_range,label= lab )
ax.vlines(x=np.linspace(0,108, num=10), ymin=0, ymax=10, color='grey', label='Octaves',linewidth=1) # plot octaves
plt.legend()
plt.show()
# [[[[[[[[[[[[[[[[[[[ -- Functions for Meteo Transformation -- ]]]]]]]]]]]]]]]]]]]
# [[[[[[[[[[[[[[[[[[[ -- Functions for Sound generation -- ]]]]]]]]]]]]]]]]]]]
import subprocess
default_soundfont = '/usr/share/sounds/sf3/MuseScore_General.sf3'
def midi_play(midi_in, soundfont= default_soundfont):
subprocess.call(['cvlc', midi_in , 'vlc://quit', '--soundfont', '/home/viturin/-vitis/Documents/MuseScore2/Soundfonts/Compifont_13082016.sf2']) # cvlc = vlc without gui
def midi_audio(midi_in, name_out = 'none', soundfont= default_soundfont):
if name_out == 'none' :
name_out = midi_in.replace('.mid', '.flac')
else:
name_out = name_out + '.flac'
subprocess.call(['mscore', '-o', name_out, midi_in]) # -o = export as
def midi_png(midi_in, name_out = 'none'):
if name_out == 'none' :
name_out = midi_in.replace('.mid', '.png')
else:
name_out = name_out + '.png'
subprocess.call(['mscore', '-o', name_out, '-T', '2', midi_in]) # -o = export as , -T 2 = cut page with 2 pixel