-
Notifications
You must be signed in to change notification settings - Fork 12
/
fm.py
422 lines (379 loc) · 16.9 KB
/
fm.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
# fm.py
# Some code to try to convert DX7 patches into AMY commands
import amy
import numpy as np
import time
from dataclasses import dataclass
from typing import List
@dataclass
class DX7Operator:
"""Per-operator parameters for DX7 patches."""
opnum: int = 0
rates: List[int] = None # 4
levels: List[int] = None # 4
breakpoint: int = 0
bp_depths: List[int] = None # 2
bp_curves: List[int] = None # 2
kbdratescaling: int = 0
ampmodsens: int = 0
keyvelsens: int = 0
ratiotuning: bool = False
freq_coarse: int = 0
freq_fine: int = 0
freq_detune: int = 0
opamp: int = 0
@dataclass
class DX7Patch:
"""Encapsulates information in a DX7 Patch."""
ops: List[DX7Operator] = None
pitch_rates: List[int] = None # 4
pitch_levels: List[int] = None # 4
algo: int = 0 # 1-32
feedback: int = 0
oscsync: int = 0
lfospeed: int = 0
lfodelay: int = 0
lfopitchmoddepth: int = 0
lfoampmoddepth: int = 0
lfosync: int = 0
lfowaveform: int = 0
pitchmodsens: int = 0
transpose: int = 0
name: str = ""
@staticmethod
def from_patch_number(patch_number):
# returns a patch (as in patches.h) from
# default-dx7-patches.bin generated by dx7db, see https://github.com/bwhitman/learnfm
f = bytes(open("default-dx7-patches.bin", mode="rb").read())
patch_data = f[patch_number*156:patch_number*156+156]
return DX7Patch.from_bytestream(bytearray(patch_data))
@staticmethod
def from_bytestream(bytestream):
"""Simply reformat the bytestream into parameters."""
result = DX7Patch()
bytestream = bytes(bytestream)
byteno = 0
def nextbyte(count=1):
nonlocal byteno
if count > 1:
# Return a list.
return [nextbyte() for _ in range(count)]
b = bytestream[byteno]
byteno += 1
# Return a bare byte.
return b
ops = []
# Starts at op 6
for i in range(6, 0, -1):
op = DX7Operator(opnum=i)
op.rates = nextbyte(4)
op.levels = nextbyte(4)
op.breakpoint = nextbyte()
op.bp_depths = nextbyte(2)
op.bp_curves = nextbyte(2)
op.kbdratescaling = nextbyte()
op.ampmodsens = nextbyte()
op.keyvelsens = nextbyte()
op.opamp = nextbyte()
op.ratiotuning = False if nextbyte() == 1 else True
op.freq_coarse = nextbyte()
op.freq_fine = nextbyte()
op.freq_detune = nextbyte()
ops.append(op)
result.ops = ops
result.pitch_rates = nextbyte(4)
result.pitch_levels = nextbyte(4)
result.algo = 1 + nextbyte()
result.feedback = nextbyte()
result.oscsync = nextbyte()
result.lfospeed = nextbyte()
result.lfodelay = nextbyte()
result.lfopitchmoddepth = nextbyte()
result.lfoampmoddepth = nextbyte()
result.lfosync = nextbyte()
result.lfowaveform = nextbyte()
result.pitchmodsens = nextbyte()
result.transpose = nextbyte()
result.name = ''.join(chr(i) for i in nextbyte(10))
return result
def get_bytestream(self):
"""Convert a decoded patch dict back to a bytestream."""
bytestream = []
for op in self.ops:
# Assume ordering is right in ops list.
bytestream.extend(op.rates)
bytestream.extend(op.levels)
bytestream.append(op.breakpoint)
bytestream.extend(op.bp_depths)
bytestream.extend(op.bp_curves)
bytestream.append(op.kbdratescaling)
bytestream.append(op.ampmodsens)
bytestream.append(op.keyvelsens)
bytestream.append(op.opamp)
bytestream.append(0 if op.ratiotuning else 1)
bytestream.append(op.freq_coarse)
bytestream.append(op.freq_fine)
bytestream.append(op.freq_detune)
bytestream.extend(self.pitch_rates)
bytestream.extend(self.pitch_levels)
bytestream.append(self.algo - 1)
bytestream.append(self.feedback)
bytestream.append(self.oscsync)
bytestream.append(self.lfospeed)
bytestream.append(self.lfodelay)
bytestream.append(self.lfopitchmoddepth)
bytestream.append(self.lfoampmoddepth)
bytestream.append(self.lfosync)
bytestream.append(self.lfowaveform)
bytestream.append(self.pitchmodsens)
bytestream.append(self.transpose)
bytestream.extend(ord(c) for c in self.name)
return bytes(bytestream)
@dataclass
class AMYOscillator:
op_num: int = 0
amp_levels: List[float] = None
amp_times: List[float] = None
op_amp: float = 0
ampmodsens: float = 0
frequency: float = 0
freq_is_ratio: bool = False
@staticmethod
def from_dx7_op(op):
result = AMYOscillator()
result.op_num = op.opnum
result.amp_levels, result.amp_times = eg_to_bp(op.rates, op.levels)
result.op_amp = 2 * dx7level_to_linear(op.opamp)
if op.ratiotuning:
result.frequency = coarse_fine_ratio(op.freq_coarse, op.freq_fine, op.freq_detune)
result.freq_is_ratio = True
else:
result.frequency = coarse_fine_fixed_hz(op.freq_coarse, op.freq_fine, op.freq_detune)
result.freq_is_ratio = False
result.ampmodsens = float(op.ampmodsens) # Don't know scaling, just 0/nonzero.
return result
def fm_trunc(number):
if(type(number)==float or type(number)==np.float64):
return ('%.6f' % number).rstrip('0').rstrip('.')
return str(number)
@dataclass
class AMYPatch:
oscs: List[AMYOscillator] = None
pitch_levels: List[float] = None
pitch_times: List[float] = None
algo: int = 0
feedback: float = 0
lfo_freq: float = 0
lfo_delay: float = 0
lfo_pitchmoddepth: float = 0
lfo_ampmoddepth: float = 0
lfo_waveform: int = 0
name: str = ""
amp_lfo_amp: float = 0
pitch_lfo_amp: float = 0
@staticmethod
def from_dx7(dx7_patch):
result = AMYPatch()
result.oscs = []
for op in dx7_patch.ops:
result.oscs.append(AMYOscillator.from_dx7_op(op))
result.pitch_levels, result.pitch_times = eg_to_bp_pitch(
dx7_patch.pitch_rates, dx7_patch.pitch_levels)
result.algo = dx7_patch.algo
result.feedback = 0.00125 * (2 ** dx7_patch.feedback)
result.lfo_freq = lfo_speed_to_hz(dx7_patch.lfospeed)
result.lfo_delay = dx7_patch.lfodelay
result.lfo_pitchmoddepth = dx7_patch.lfopitchmoddepth
result.lfo_ampmoddepth = dx7_patch.lfoampmoddepth
result.lfo_waveform = lfo_wave(dx7_patch.lfowaveform)
result.amp_lfo_amp = dx7level_to_linear(result.lfo_ampmoddepth)
result.pitch_lfo_amp = dx7level_to_linear(result.lfo_pitchmoddepth)
result.name = dx7_patch.name
return result
def send_to_AMY(self, reset=True):
# Take a FM patch and output AMY commands to set up the patch.
# Send amy.send(vel=1,osc=0,note=50) after
t = fm_trunc
if(reset): amy.reset()
pitch_levels, pitch_times = self.pitch_levels, self.pitch_times
pitchbp = "%d,%s,%d,%s,%d,%s,%d,%s,%d,%s" % (
pitch_times[0], t(pitch_levels[0]), pitch_times[1], t(pitch_levels[1]),
pitch_times[2], t(pitch_levels[2]), pitch_times[3], t(pitch_levels[3]),
pitch_times[4], t(pitch_levels[4]))
# Set up each operator.
last_release_time = 0
last_release_value = 0
for i, osc in enumerate(self.oscs):
amp_levels, amp_times = osc.amp_levels, osc.amp_times
oscbp = "%d,%s,%d,%s,%d,%s,%d,%s,%d,%s" % (
amp_times[0], t(amp_levels[0]), amp_times[1], t(amp_levels[1]),
amp_times[2], t(amp_levels[2]), amp_times[3], t(amp_levels[3]),
amp_times[4], t(amp_levels[4]))
oscbpfmt = "%d,%s/%d,%s/%d,%s/%d,%s/%d,%s" % (
amp_times[0], t(amp_levels[0]), amp_times[1], t(amp_levels[1]),
amp_times[2], t(amp_levels[2]), amp_times[3], t(amp_levels[3]),
amp_times[4], t(amp_levels[4]))
if(amp_times[4] > last_release_time):
last_release_time = amp_times[4]
last_release_value = amp_levels[4]
#print("osc %d (op %d) freq %.6f ratio %d env %s amp %.6f amp_mod %d" % \
# (i+1, osc.op_num, osc.frequency, osc.freq_is_ratio, oscbpfmt,
# osc.op_amp, osc.ampmodsens))
# Make them all in cosine phase, to be like DX7. Important for slow oscs
args = {"osc": i + 1,
"bp0": oscbp, "phase": 0.25}
if osc.freq_is_ratio:
args["ratio"] = t(osc.frequency)
else:
args["freq"] = t(osc.frequency)
# TODO: we ignore intensity of amp mod sens, just on/off
args.update({"mod_source": 7, "amp": "%s,0,0,1,0,%d" % (t(osc.op_amp), osc.ampmodsens > 0)})
# We are _NOT_ updating operators with pitch bp, per dan tuesday 7/5 morning (but not monday 7/4 morning)
#args.update({"bp1": pitchbp})
amy.send(**args)
# Set up the amp LFO
#print("osc 7 amp lfo wave %d freq %f amp %f" % (
# self.lfo_waveform, self.lfo_freq, self.amp_lfo_amp))
amy.send(osc=7, wave=self.lfo_waveform, freq=t(self.lfo_freq),
amp=t(self.amp_lfo_amp))
# and the pitch one
#print("osc 8 pitch lfo wave %d freq %f amp %f" % (
# self.lfo_waveform, self.lfo_freq, self.pitch_lfo_amp))
amy.send(osc=8, wave=self.lfo_waveform, freq=t(self.lfo_freq),
amp=t(self.pitch_lfo_amp))
#print("not used: lfo delay %d " % self.lfo_delay)
ampbp = "0,1,%d,%f" % (last_release_time, last_release_value)
#print("osc 0 (main) algo %d feedback %f pitchenv %s ampenv %s" % (
# self.algo, self.feedback, pitchbp, ampbp))
amy.send(osc=0, wave=amy.ALGO, algorithm=self.algo, feedback=t(self.feedback),
algo_source="1,2,3,4,5,6",
bp0=ampbp,
bp1=pitchbp,
freq="0,1,0,0,1,1", mod_source=8)
def dx7level_to_linear(dx7level):
"""Map the dx7 0..99 levels to linear amplitude."""
return 2 ** ((dx7level - 99) / 8)
def linear_to_dx7level(linear):
"""Map a linear amplitude to the dx7 0..99 scale."""
return np.log2(np.maximum(dx7level_to_linear(0), linear)) * 8 + 99
def pitchval_to_ratio(pitchval):
"""Map 0..99 DX7 pitch vals (e.g. from pitch_env) into f0 ratios."""
# Pitch map 0..99 actually becomes -128..127 via a symmetric map with 50->0, linear from 15 to 85, then
# quadratic in the remainder.
pitchsign = -1 + 2*(pitchval >= 50)
semipitchval = np.abs(pitchval - 50).astype(float)
# Above (50 + 36), Quadratic to reach 127 at level 99.
semipitchval += (semipitchval > 36) * (((semipitchval - 34)**2) * 93/225 - semipitchval + 34)
# DX7 manual states pitchmod range is +/- 4 octaves, so 32 steps/oct sounds right.
return 2 ** ((pitchsign * semipitchval) / 32)
def ratio_to_pitchval(ratio):
semipitchval = 32 * np.log2(ratio)
pitchsign = -1 + 2*(semipitchval >= 0)
semipitchval = np.abs(semipitchval)
# Vectorized conditional treatment of outside -36 to 36.
semipitchval += (semipitchval > 36) * (34 + np.sqrt(np.abs(semipitchval - 34) * (225/93)) - semipitchval)
return 50 + pitchsign * semipitchval
def calc_loglin_eg_breakpoints(rates, levels, dx7_attacks=True,
rate_double_interval=6, rate_scale=0.5, rate_offset=0.5):
"""Convert the DX7 rates/levels into (time, target) pairs (for amy)"""
if dx7_attacks:
level_to_lin_fn = dx7level_to_linear
else:
level_to_lin_fn = pitchval_to_ratio
# This is the part we precompute in fm.py to get breakpoints to send to amy.
current_level = levels[-1]
# EG at time 0 has final value from release.
breakpoints = [(0, level_to_lin_fn(current_level))]
MIN_LEVEL = 34
ATTACK_RANGE = 75
def level_to_attack_time(level, t_const):
"""Return the time at which a paradigmatic DX7 attack envelope will reach a level (0..99 range)"""
# Return the t0 that solves level = MIN_LEVEL + ATTACK_RANGE * (1 - exp(-t0 / t_const))
return -t_const * np.log((MIN_LEVEL + ATTACK_RANGE - np.maximum(MIN_LEVEL, level))/ATTACK_RANGE)
for segment, (rate, target_level) in enumerate(zip(rates, levels)):
release_segment = (segment == len(rates)-1)
if dx7_attacks and target_level > current_level: # Attack segment
# The attack envelopes L(t) appear to be ~ 34 + 75 * (1 - exp(t / t_const)), starting from L = 34
# i.e. they are rising exponentials (as in analog ADSR, but here in the log(amp) domain)
# with an asymptote at 109 (i.e., 10 higher than the highest possible amp).
# The time constant depends on the R (rate) parameter, and is well fit by:
t_const = 0.008 * (2 ** ((65 - rate)/6))
# Total time for this segment is t1 - t0 where t0 and t1 solve
# effective_start = 34 + 75 * (1 - np.exp(-t0 / t_const)) = 109 - 75 exp(-t0 / t_c)
# target_level = 34 + 75 * (1 - np.exp(-t1 / t_const)) = 109 - 75 exp(-t1 / t_c)
# so t1 - t0 = -t_c * [log((34 + 75 - target_level)/75) - log((34 + 75 - effective_start)/75)]
effective_start_level = np.maximum(current_level, MIN_LEVEL)
t0 = level_to_attack_time(effective_start_level, t_const)
segment_duration = level_to_attack_time(target_level, t_const) - t0
#print("eff_st=", effective_start_level, "t_c=", t_const, "t0=", t0, "dur=", segment_duration)
# Now amy's task will be to recover t0 and t_const from (time, target) pairs
else:
# Decay segment, or TRUE_EXPONENTIAL attack segment.
direction = 1 if target_level > current_level else -1
# "A falling segment takes 3.5 mins"
# so delta = 99 in 210 seconds -> level_change_per_sec = 0.5
# I think just offset everything by 0.5, avoids div0.
level_change_per_sec = direction*(rate_offset + rate_scale * (2 ** (rate / rate_double_interval)))
level_difference = target_level - current_level
# Hack to cover for sustain = 0, release = 0 release segments which look like they should be zero long
if release_segment and level_difference == 0:
level_difference = direction * 60 # e.g. from a decayed level of 80 to zero.
#print("** Goosing release amp")
segment_duration = level_difference / level_change_per_sec
#print("lcps=", level_change_per_sec, "dur=", segment_duration)
breakpoints.append((segment_duration, level_to_lin_fn(target_level)))
current_level = target_level
return breakpoints
def eg_to_bp(egrate, eglevel, calc_eg_args={}):
breakpoints = calc_loglin_eg_breakpoints(egrate, eglevel, **calc_eg_args)
rates = []
times = []
for time, level in breakpoints:
times.append(int(1000 * time))
rates.append(level)
return rates, times
def eg_to_bp_pitch(egrate, eglevel):
# Additional args to make breakpoint calculation to the right thing for pitch.
calc_pitch_eg_args = {'dx7_attacks': False, 'rate_double_interval': 20, 'rate_scale': 11, 'rate_offset': -6}
return eg_to_bp(egrate, eglevel, calc_pitch_eg_args)
def coarse_fine_fixed_hz(coarse, fine, detune=7):
coarse = coarse & 3
return 10 ** (coarse + (fine + ((detune - 7) / 8)) / 100 )
def coarse_fine_ratio(coarse, fine, detune=7):
coarse = coarse & 31
if(coarse == 0):
coarse = 0.5
return coarse * (1 + (fine + ((detune - 7) / 8)) / 100)
def lfo_speed_to_hz(byte):
# Measured values from TX802, linear fit by eye
if byte == 0:
return 0.064
if byte <= 64:
return byte / 6.0
if byte <= 85:
return byte - 64.0 * 5.0/6.0
# Byte > 85
return 31.67 + (byte - 85.0) * 1.33
def lfo_wave(byte):
if byte > 5:
return None
return [
amy.TRIANGLE, amy.SAW_DOWN, amy.SAW_UP,
amy.PULSE, amy.SINE, amy.NOISE
][byte]
# Play a numpy array on an Apple Silicon mac without having to use an external library
# (sounddevice is currently broken on AS macs)
def play_np_array(np_array, samplerate=amy.AMY_SAMPLE_RATE):
import wave, tempfile , os, struct
tf = tempfile.NamedTemporaryFile()
obj = wave.open(tf,'wb')
obj.setnchannels(1) # mono
obj.setsampwidth(2)
obj.setframerate(samplerate)
for i in range(np_array.shape[0]):
value = int(np_array[i] * 32767.0)
data = struct.pack('<h', value)
obj.writeframesraw( data )
obj.close()
os.system("afplay " + tf.name)
tf.close()