def setup(self):
if self.pwm:
self.io.setMode(self.pin)
self.io.write(self.pin, 0)
else:
self.dac = DAC(self.io)
self.dac.setup()
def __init__(self, namespace):
Tk.__init__(self, className="Koshka")
self.dac = DAC()
self.dac.start()
self.score_path = namespace.score
self.sequencer = None
self.sequencer_frame = None
self.mixer_window = None
self.scale_window = None
self._open_score(self.score_path)
menu = Menu(self)
self.config(menu=menu)
filemenu = Menu(menu)
menu.add_cascade(label="File", menu=filemenu)
filemenu.add_command(label="Open...", command=self.open, accelerator="meta-o")
filemenu.add_command(label="Save", command=self.save, accelerator="meta-s")
filemenu.add_command(label="Save As...", command=self.save_as, accelerator="meta-shift-s")
filemenu.add_separator()
filemenu.add_command(label="Exit", command=self.quit)
menu.add_cascade(label="Help", menu=filemenu)
filemenu.add_command(label="Online Help...", command=lambda: webbrowser.open_new_tab(URL_HELP_DOC))
# Note: This is only implemented and tested for Mac OS
self.bind_all("<Command-o>", self.open)
self.bind_all("<Command-s>", self.save)
self.bind_all("<Command-Shift-s>", self.save_as)
self.bind_all("<Meta-o>", self.open)
self.bind_all("<Meta-s>", self.save)
self.bind_all("<Meta-Shift-s>", self.save_as)
def dac_test():
tb = DAC()
cd_dac = ClockDomain("dac", reset_less=True)
tb.clock_domains += cd_dac
def run(tb):
ida = 0x28FC
qda = 0x3CFC
dut = tb
for i in range(12):
yield dut.idata.eq(ida)
yield dut.qdata.eq(qda)
yield
yield dut.idata.eq(ida + 10)
yield dut.qdata.eq(qda - 20)
yield
yield dut.idata.eq(ida + 20)
yield dut.qdata.eq(qda - 30)
yield
for i in range(20):
yield
run_simulation(tb, run(tb),
vcd_name="dac_test.vcd",
clocks = {
"sys": (8, 0),
"dac": (4, 0),
},
)
def train_dac_model(self, model_params):
dac = DAC()
smoother_model = dac.build_model(hidden_sizes=[64, 64],
seq_len=50,
no_words=40000,
emb_layer=self.embedding_layer,
lr=0.01)
generator = Generator(sequences=self.sequences,
batch_size=SMOOTH_BS,
max_words=MAX_NUM_WORDS,
max_len=MAX_LEN,
split=SMOOTH_SPLIT)
smoother_model = dac.train(generator,
full_model=smoother_model,
model_params=model_params,
bs=SMOOTH_BS,
split=SMOOTH_SPLIT,
pretrain_epochs=4,
epochs=SMOOTH_EPOCHS)
class TestSampler(unittest.TestCase):
def setUp(self):
self.dac = DAC(BUFFER_SIZE, SAMPLE_RATE)
self.dac.start()
def tearDown(self):
self.dac.stop()
def test_instantiation(self):
sample = SingleSoundSampler(SAMPLE_FILE)
self.assertIsNotNone(sample)
def test_polyphonic_play(self):
sampler = Sampler()
sampler.add_sample(0, SAMPLE_FILE)
sampler.add_sample(1, SAMPLE_FILE)
self.dac.connect(sampler.callback)
sampler.on(0)
sleep(0.5)
sampler.on(0)
sleep(0.5)
sampler.on(1)
sleep(2)
sampler.off(0)
sampler.off(1)
class TestInstrument(unittest.TestCase):
def setUp(self):
self.scale = PythagSeries(BASE_FREQUENCY)
self.dac = DAC(BUFFER_SIZE, SAMPLE_RATE)
self.dac.start()
def tearDown(self):
self.dac.stop()
def test_instantiation(self):
instrument = MonophonicScaleSynth(SAMPLE_RATE, BUFFER_SIZE, self.scale)
self.assertIsNotNone(instrument)
def test_play(self):
instrument = ScaleSynth(SAMPLE_RATE, BUFFER_SIZE, self.scale)
# instrument = ScaleSynth(SAMPLE_RATE, BUFFER_SIZE, self.scale)
self.dac.connect(instrument.callback)
instrument.on(0)
sleep(.5)
instrument.off(0)
sleep(.1)
instrument.on(1)
instrument.on(4)
sleep(.5)
instrument.off(1)
sleep(.5)
def run(self, predict_path, smoother_path, eval_path):
self._parse_corpus(MIN_SEQ_LEN, TEXT_DATA_DIR + TEXT_DATA)
self.prepare_emb(EMBEDDING_DIM, MAX_LEN)
predict_model = None
if predict_path is None:
model_params = {
'lstm': [16],
'merge_layer': 'concat',
'dense': {
'size': [64, 32],
'act': 'elu',
'dropout': 0
},
'optimizer': 'adam',
'lr': 0.0005
}
predict_model = self.train_predict_model(model_params)
else:
pass
#predict_model = self.load_predict_model(predict_path)
#smoother_model = None
if smoother_path is None:
model_params = {'size': [64, 64], 'lr': 0.01}
#smoother_model = self.train_dac_model(model_params)
else:
self.dac = DAC()
self.dac.load_model(smoother_path)
#GENERATE PUN
while True:
try:
final = pungen.form_pun(eval_path)
break
except Exception:
pass
print(final)
class Throttle():
vref = 5
def __init__(self, io, pin, pwm):
'''exit
pwm = 0 for DAC output
pwm = 1 for PWM output
'''
self.io = io
self.pin = pin
self.pwm = pwm
self.maxV = 3.7
self.minV = 0.0
def setup(self):
if self.pwm:
self.io.setMode(self.pin)
self.io.write(self.pin, 0)
else:
self.dac = DAC(self.io)
self.dac.setup()
def setVolt(self, volt):
if volt < self.minV or volt > self.maxV:
raise ThrottleError('Volt out of bound: ', volt, ' V')
if self.pwm:
duty = self.calcDuty(volt)
self.pwmWrite(duty)
else:
self.dacWrite(volt)
def calcDuty(self, volt):
return int(volt/Throttle.vref * (2**8-1))
def off(self):
if self.pwm:
self.pwmWrite(0)
else:
self.dac.off()
def pwmWrite(self, duty, freq = 1000):
if duty == 0:
self.io.pwmStop(self.pin)
else:
self.io.pwmStart(self.pin, duty, freq)
def dacWrite(self, volt):
self.dac.setVolt(volt)
def setUp(self):
self.dac = DAC(BUFFER_SIZE, SAMPLE_RATE)
self.dac.start()
from dac import DAC
from sequencers.grid_sequencer import GridSequencer
DEFAULT_SCORE_FILE = "./scores/score_sample_blank.txt"
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Koshka - MSP project")
parser.add_argument("score", nargs="?", type=str, default=DEFAULT_SCORE_FILE)
parser.add_argument("--no_gui", type=bool, default=False, const=True, nargs="?")
parser.add_argument("-l", "--loop", type=int, default=0, const=GridSequencer.INFINIT_LOOP, nargs="?")
namespace = parser.parse_args()
if namespace.no_gui:
dac = DAC()
dac.start()
try:
sequencer = GridSequencer(namespace.score, buffer_size=dac.bufferSize, sample_rate=dac.getSamplerate())
dac.connect(sequencer.callback)
sequencer.play(namespace.loop)
while sequencer.running:
sleep(0.1)
finally:
dac.stop()
else:
from gui.main_window import MainWindow
def setUp(self):
self.scale = PythagSeries(BASE_FREQUENCY)
self.dac = DAC(BUFFER_SIZE, SAMPLE_RATE)
self.dac.start()
import instruments
import argparse
DEFAULT_INSTRUMENT = "ScaleSynth EvenTempered 528"
DEFAULT_SPEED = 120
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='MSP homework 9')
parser.add_argument('score', type=str)
parser.add_argument('-i', '--instrument', type=str, default=DEFAULT_INSTRUMENT)
parser.add_argument('-s', '--speed', type=int, default=DEFAULT_SPEED)
namespace = parser.parse_args()
dac = DAC(bufferSize=2 ** 10, rate=44100)
dac.start()
try:
sequencer = Sequencer(buffer_size=dac.bufferSize, sample_rate=dac.getSamplerate(), speed=namespace.speed)
dac.connect(sequencer.callback)
instrument = instruments.parse(namespace.instrument.split(),
buffer_size=dac.bufferSize,
sample_rate=dac.getSamplerate())
sequencer.add_instrument(instrument)
file = open(namespace.score, 'r')
score_string = file.read()
file.close()
class MainWindow(Tk):
def __init__(self, namespace):
Tk.__init__(self, className="Koshka")
self.dac = DAC()
self.dac.start()
self.score_path = namespace.score
self.sequencer = None
self.sequencer_frame = None
self.mixer_window = None
self.scale_window = None
self._open_score(self.score_path)
menu = Menu(self)
self.config(menu=menu)
filemenu = Menu(menu)
menu.add_cascade(label="File", menu=filemenu)
filemenu.add_command(label="Open...", command=self.open, accelerator="meta-o")
filemenu.add_command(label="Save", command=self.save, accelerator="meta-s")
filemenu.add_command(label="Save As...", command=self.save_as, accelerator="meta-shift-s")
filemenu.add_separator()
filemenu.add_command(label="Exit", command=self.quit)
menu.add_cascade(label="Help", menu=filemenu)
filemenu.add_command(label="Online Help...", command=lambda: webbrowser.open_new_tab(URL_HELP_DOC))
# Note: This is only implemented and tested for Mac OS
self.bind_all("<Command-o>", self.open)
self.bind_all("<Command-s>", self.save)
self.bind_all("<Command-Shift-s>", self.save_as)
self.bind_all("<Meta-o>", self.open)
self.bind_all("<Meta-s>", self.save)
self.bind_all("<Meta-Shift-s>", self.save_as)
# self.wm_attributes("-titlepath",'What is this?')
def __delete__(self, instance):
print "__delete__"
self._quit()
def quit(self):
print "tk quit"
Tk.quit(self)
pass
def _quit(self):
print "_quit"
self.dac.stop()
self.destroy()
self.quit()
def _open_sequencer(self, sequencer):
print "\nMainWindow: _open_sequencer"
# print("Stop DAC...")
self.dac.stop()
# print("Stop DAC...done")
if self.sequencer_frame:
self.sequencer_frame.destroy()
if self.mixer_window:
self.mixer_window.destroy()
if self.scale_window is not None:
self.scale_window.destroy()
# print("Reset sequencer...")
if self.sequencer is not None:
self.sequencer.stop()
self.sequencer.remove_all_observers()
for i in self.sequencer.instruments:
i.remove_all_observers()
# print("Reset sequencer...done")
# print("Connect sequencder...")
self.sequencer = sequencer
self.dac.connect(self.sequencer.callback)
# print("Connect sequencder...done")
for i in sequencer.instruments:
i.id_variable = StringVar()
i.id_variable.set(i.name_id)
self.sequencer_frame = SequencerFrame(self, self.sequencer)
self.sequencer_frame.pack()
self.mixer_window = mixer_gui.MixerWindow(self, self.sequencer)
self.scale_window = keyboard.ScaleWindow(self)
for i in sequencer.instruments:
self.scale_window.add_instrument(i)
self.dac.start()
pass
def _open_score(self, score_path):
self.score_path = score_path
sequencer = GridSequencer(score_path, buffer_size=self.dac.bufferSize, sample_rate=self.dac.getSamplerate())
self._open_sequencer(sequencer)
def open(self, val=Tkinter.Event()):
score_path = tkFileDialog.askopenfilename(filetypes=[("Text file", ".txt")])
if score_path:
self._open_score(score_path)
def save(self, val=None):
self.sequencer.save(self.score_path)
# self.wm_attributes("-modified", 0)
def save_as(self, val=None):
self.score_path = tkFileDialog.asksaveasfilename(filetypes=[("Text file", ".txt")])
self.save()
###############################################################################
# SCHEDULER
##########
# You shall modify speedCoeff only
###############################################################################
scheduler = Scheduler(speedCoeff=1)
###############################################################################
# REGISTERS
##########
# Add registers like that : reg = Register("NAME")
###############################################################################
###############################################################################
# ECUS
##########
# You should always add registers in same order. Else there is an overlap.
# TODO I want to find out how to add registers with no order constraint
###############################################################################
allEcus = [
Ecu("../examples/viper2/App-GBF/trampoline", scheduler, [
Timer("TIMER0", 1, type=timer.AUTO, delay=10),
DAC("DAC0", 2, position=[0, 0]),
LCD("LCD1", 3, position=[360, 0]),
LCD("LCD2", 4, position=[360, 120]),
BP("BPPlus", 10, position=[360, 240], picture="pictures/BPPlus"),
BP("BPMinus", 11, position=[460, 240], picture="pictures/BPMinus"),
Power("POWER", 9),
])
]
class Pungen:
def __init__(self, **kwargs):
self.filepath = kwargs.get('filepath')
self.embedding_layer = None
def _parse_corpus(self, min_seq_len, filepath):
print('Indexing word vectors.')
self.texts = []
with open(filepath, encoding='utf-8') as fp:
for line in fp:
if line == "\n":
continue
self.texts.append(line)
self.tokenizer = Tokenizer(num_words=MAX_NUM_WORDS,
filters=TOKEN_FILTER)
self.tokenizer.fit_on_texts(self.texts)
self.sequences = self.tokenizer.texts_to_sequences(self.texts)
self.sequences = [x for x in self.sequences if len(x) >= min_seq_len]
self.word_index = self.tokenizer.word_index
print('Found %s unique tokens.' % len(self.word_index))
print('Found %s texts.' % len(self.sequences))
def prepare_emb(self, emb_dim, input_length):
print('Indexing word vectors.')
emb_name = 'glove.6B.' + str(emb_dim) + "d.txt"
self.embeddings_index = {}
with open(os.path.join(GLOVE_DIR, emb_name), encoding='utf-8') as f:
for line in f:
word, coefs = line.split(maxsplit=1)
coefs = np.fromstring(coefs, 'f', sep=' ')
self.embeddings_index[word] = coefs
print('Found %s word vectors.' % len(self.embeddings_index))
# prepare embedding matrix
num_words = MAX_NUM_WORDS
self.embedding_matrix = np.zeros((num_words, emb_dim))
for word, i in self.word_index.items():
if i >= num_words:
continue
embedding_vector = self.embeddings_index.get(word)
if embedding_vector is not None:
# words not found in embedding index will be all-zeros.
self.embedding_matrix[i] = embedding_vector
# load pre-trained word embeddings into an Embedding layer
# note that we set trainable = False so as to keep the embeddings fixed
self.embedding_layer = Embedding(num_words,
emb_dim,
embeddings_initializer=Constant(
self.embedding_matrix),
input_length=input_length,
trainable=False)
def check_generator(self):
texts = self.tokenizer.sequences_to_texts(self.sequences)
if len(texts) != len(self.texts):
print("Different sizes of texts")
return
filter = set(TOKEN_FILTER)
for i in range(len(texts)):
if texts[i].lower() != self.texts[i][:-1].lower():
if any((c in filter) for c in self.texts[i][:-1].lower()):
continue
print(texts[i], self.texts[i][:-1])
print(self.texts[i][:-1].lower())
print("Tokenizer failed to tokenize properly!")
return
print("Tokenizer check was succesfull!")
def form_pun(self, eval_path):
retrieve = Retrieve(sentence_path=TEXT_DATA_DIR + TEXT_DATA,
pun_path=PUN_DATA_DIR + PUN_DATA)
(pun, sentence, score) = retrieve.retrieve()
if not sentence:
print("No sentence with word {} was found. Exiting...".format(
pun[1]))
raise Exception()
text = word_tokenize(sentence)
tokenized = nltk.pos_tag(text)
print(tokenized)
print(sentence, pun[0], pun[1])
pre = self.tokenizer.texts_to_sequences([sentence])
wp = self.tokenizer.texts_to_sequences([pun[0]])
wa = self.tokenizer.texts_to_sequences([pun[1]])
if (not wa[0]) or (not wp[0]):
print(
"The pair of pun and word does not exist in the parsed corpus. Exit..."
)
raise Exception()
index_wa = -1
for seq in pre[0]:
index_wa = index_wa + 1
if seq == wa[0][0]:
pre[0][index_wa] = wp[0][0]
break
wordsimilarity = WordSimilarity()
wordsimilarity.word2vec()
wordsimilarity.load()
try_limit = 5
try_count = 0
index_topic = 0
while True:
try:
topic_word = None
for i in range(index_topic, len(tokenized)):
(word, pos) = tokenized[i]
if (pos == 'NNP'):
topic_word = "man"
print(word, pos)
index_topic = index_topic + 1
break
if (pos == 'NN') or (pos == 'PRP') or (pos == 'NNS') or (
pos == 'PRP$'):
topic_word = word
print(word, pos)
index_topic = index_topic + 1
break
index_topic = index_topic + 1
result = wordsimilarity.getSimilar([topic_word, pun[0]],
[pun[1]], 10)
other_result = wordsimilarity.getSimilar([pun[0]], [], 10)
break
except KeyError:
print("Word {} is not in vocabulary, try with the next one".
format(topic_word))
try_count = try_count + 1
if try_limit == try_count:
print("Limit of trys has been reached. Exit...")
raise Exception()
eval_surprisal = Evaluate()
eval_surprisal.load_model(eval_path)
finals = []
mean_amalgam = 0
for (word, prob) in result:
swap = self.tokenizer.texts_to_sequences([word])
context_window = 2
surprise = eval_surprisal.compute_surpisal(
sentence=pre[0],
pun_word=wa[0][0],
pun_alternative=wp[0][0],
context_window=context_window)
mean_amalgam = mean_amalgam + surprise
print(surprise)
pre[0][index_topic] = swap[0][0]
post_simple = self.tokenizer.sequences_to_texts([pre[0]])
print(post_simple)
pre[0][index_topic + 1] = 0
if index_topic >= 2:
pre[0][index_topic - 1] = 0
post_smoothing = self.dac.inference(pre[0])
post_smoothing = self.tokenizer.sequences_to_texts(
post_smoothing.tolist())
finals.append(post_smoothing)
print(post_smoothing)
print(finals)
print(mean_amalgam / 10)
other_finals = []
mean_similar = 0
for (word, prob) in other_result:
swap = self.tokenizer.texts_to_sequences([word])
context_window = 2
surprise = eval_surprisal.compute_surpisal(
sentence=pre[0],
pun_word=wa[0][0],
pun_alternative=wp[0][0],
context_window=context_window)
mean_similar = mean_similar + surprise
print(surprise)
pre[0][index_topic] = swap[0][0]
post_simple = self.tokenizer.sequences_to_texts([pre[0]])
print(post_simple)
pre[0][index_topic + 1] = 0
if index_topic >= 2:
pre[0][index_topic - 1] = 0
post_smoothing = self.dac.inference(pre[0])
post_smoothing = self.tokenizer.sequences_to_texts(
post_smoothing.tolist())
other_finals.append(post_smoothing)
print(post_smoothing)
print(other_finals)
print(mean_similar / 10)
return finals.extend(other_finals)
def train_predict_model(self, model_params):
predict_word = WordPredict(max_len=MAX_LEN,
max_words=MAX_NUM_WORDS,
emb_layer=self.embedding_layer)
predict_word.build_model(**model_params)
predict_word.compile_model(model_params)
generator = Generator(sequences=self.sequences,
batch_size=PREDICT_BS,
max_words=MAX_NUM_WORDS,
max_len=MAX_LEN,
split=PREDICT_SPLIT)
predict_word.train(generator, PREDICT_BS, PREDICT_SPLIT,
PREDICT_EPOCHS)
return predict_word
def load_predict_model(self, path):
predict_word = load_model(path)
return predict_word
def train_dac_model(self, model_params):
dac = DAC()
smoother_model = dac.build_model(hidden_sizes=[64, 64],
seq_len=50,
no_words=40000,
emb_layer=self.embedding_layer,
lr=0.01)
generator = Generator(sequences=self.sequences,
batch_size=SMOOTH_BS,
max_words=MAX_NUM_WORDS,
max_len=MAX_LEN,
split=SMOOTH_SPLIT)
smoother_model = dac.train(generator,
full_model=smoother_model,
model_params=model_params,
bs=SMOOTH_BS,
split=SMOOTH_SPLIT,
pretrain_epochs=4,
epochs=SMOOTH_EPOCHS)
def run(self, predict_path, smoother_path, eval_path):
self._parse_corpus(MIN_SEQ_LEN, TEXT_DATA_DIR + TEXT_DATA)
self.prepare_emb(EMBEDDING_DIM, MAX_LEN)
predict_model = None
if predict_path is None:
model_params = {
'lstm': [16],
'merge_layer': 'concat',
'dense': {
'size': [64, 32],
'act': 'elu',
'dropout': 0
},
'optimizer': 'adam',
'lr': 0.0005
}
predict_model = self.train_predict_model(model_params)
else:
pass
#predict_model = self.load_predict_model(predict_path)
#smoother_model = None
if smoother_path is None:
model_params = {'size': [64, 64], 'lr': 0.01}
#smoother_model = self.train_dac_model(model_params)
else:
self.dac = DAC()
self.dac.load_model(smoother_path)
#GENERATE PUN
while True:
try:
final = pungen.form_pun(eval_path)
break
except Exception:
pass
print(final)