def init_motor_DMP(self,
optim_initial_position=True,
optim_end_position=True,
default_motor_initial_position=None,
default_motor_end_position=None):
default = np.zeros(self.n_dynamic_motor_dims * (self.n_bfs + 2))
if not optim_initial_position:
default[:self.
n_dynamic_motor_dims] = default_motor_initial_position or [
0.
] * self.n_dynamic_motor_dims
dims_optim = [False] * self.n_dynamic_motor_dims
else:
dims_optim = [True] * self.n_dynamic_motor_dims
dims_optim += [True] * (self.n_dynamic_motor_dims * self.n_bfs)
if not optim_end_position:
default[-self.
n_dynamic_motor_dims:] = default_motor_end_position or [
0.
] * self.n_dynamic_motor_dims
dims_optim += [False] * self.n_dynamic_motor_dims
else:
dims_optim += [True] * self.n_dynamic_motor_dims
self.motor_dmp = DmpPrimitive(self.n_dynamic_motor_dims,
self.n_bfs,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def __init__(self,
n_dmps,
n_bfs,
used,
default,
conf,
sm,
im,
dmp_type='discrete',
ay=None):
Agent.__init__(self, conf, sm, im)
self.n_dmps, self.n_bfs = n_dmps, n_bfs
self.current_m = zeros(self.conf.m_ndims)
if ay is None:
self.dmp = DmpPrimitive(n_dmps,
n_bfs,
used,
default,
type=dmp_type)
else:
self.dmp = DmpPrimitive(n_dmps,
n_bfs,
used,
default,
type=dmp_type,
ay=ones(n_dmps) * 1.)
def __init__(self,
synth,
m_mins,
m_maxs,
s_mins,
s_maxs,
m_used,
s_used,
n_dmps,
n_bfs,
dmp_move_steps,
dmp_max_param,
sensory_traj_samples,
audio,
diva_path=None):
self.m_mins = m_mins
self.m_maxs = m_maxs
self.s_mins = s_mins
self.s_maxs = s_maxs
self.m_used = m_used
self.s_used = s_used
self.n_dmps = n_dmps
self.n_bfs = n_bfs
self.dmp_move_steps = dmp_move_steps
self.dmp_max_param = dmp_max_param
self.samples = array(linspace(-1, self.dmp_move_steps - 1,
sensory_traj_samples + 1),
dtype=int)[1:]
self.audio = audio
self.f0 = 1.
self.pressure = 1.
self.voicing = 1.
if self.audio:
import pyaudio
self.pa = pyaudio.PyAudio()
self.stream = self.pa.open(format=pyaudio.paFloat32,
channels=1,
rate=11025,
output=True)
if synth == "octave":
self.synth = DivaOctaveSynth(diva_path)
elif synth == "matlab":
self.synth = DivaMatlabSynth()
else:
raise NotImplementedError
self.art = array(
[0.] * 10 + [self.f0, self.pressure, self.voicing]
) # 13 articulators is a constant from diva_synth.m in the diva source code
self.max_params = array([1.] * self.n_dmps + [self.dmp_max_param] *
self.n_bfs * self.n_dmps + [1.] * self.n_dmps)
self.dmp = DmpPrimitive(dmps=self.n_dmps,
bfs=self.n_bfs,
timesteps=self.dmp_move_steps)
Environment.__init__(self, self.m_mins, self.m_maxs, self.s_mins,
self.s_maxs)
def init_sensori_DMP(self,
optim_initial_position=True,
optim_end_position=True,
default_sensori_initial_position=None,
default_sensori_end_position=None):
default = np.zeros(self.n_dynamic_sensori_dims *
(self.n_sensori_traj_points + 2))
if not optim_initial_position:
default[:self.
n_dynamic_sensori_dims] = default_sensori_initial_position
dims_optim = [False] * self.n_dynamic_sensori_dims
else:
dims_optim = [True] * self.n_dynamic_sensori_dims
dims_optim += [True] * (self.n_dynamic_sensori_dims *
self.n_sensori_traj_points)
if not optim_end_position:
default[-self.
n_dynamic_sensori_dims:] = default_sensori_end_position
dims_optim += [False] * self.n_dynamic_sensori_dims
else:
dims_optim += [True] * self.n_dynamic_sensori_dims
self.sensori_dmp = DmpPrimitive(self.n_dynamic_sensori_dims,
self.n_sensori_traj_points,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def __init__(self,
synth,
m_mins, m_maxs, s_mins, s_maxs,
m_used,
s_used,
n_dmps,
n_bfs,
dmp_move_steps,
dmp_max_param,
sensory_traj_samples,
audio,
diva_path=None):
self.m_mins = m_mins
self.m_maxs = m_maxs
self.s_mins = s_mins
self.s_maxs = s_maxs
self.m_used = m_used
self.s_used = s_used
self.n_dmps = n_dmps
self.n_bfs = n_bfs
self.dmp_move_steps = dmp_move_steps
self.dmp_max_param = dmp_max_param
self.samples = array(linspace(-1, self.dmp_move_steps-1, sensory_traj_samples + 1), dtype=int)[1:]
self.audio = audio
self.f0 = 1.
self.pressure = 1.
self.voicing = 1.
if self.audio:
import pyaudio
self.pa = pyaudio.PyAudio()
self.stream = self.pa.open(format=pyaudio.paFloat32,
channels=1,
rate=11025,
output=True)
if synth == "octave":
self.synth = DivaOctaveSynth(diva_path)
elif synth == "matlab":
self.synth = DivaMatlabSynth()
else:
raise NotImplementedError
self.art = array([0.]*10 + [self.f0, self.pressure, self.voicing]) # 13 articulators is a constant from diva_synth.m in the diva source code
self.max_params = array([1.] * self.n_dmps + [self.dmp_max_param] * self.n_bfs * self.n_dmps + [1.] * self.n_dmps)
self.dmp = DmpPrimitive(dmps=self.n_dmps, bfs=self.n_bfs, timesteps=self.dmp_move_steps)
Environment.__init__(self, self.m_mins, self.m_maxs, self.s_mins, self.s_maxs)
def init_motor_DMP(self, optim_initial_position=True, optim_end_position=True, default_motor_initial_position=None, default_motor_end_position=None):
default = np.zeros(self.n_dynamic_motor_dims * (self.n_bfs + 2))
if not optim_initial_position:
default[:self.n_dynamic_motor_dims] = default_motor_initial_position or [0.] * self.n_dynamic_motor_dims
dims_optim = [False] * self.n_dynamic_motor_dims
else:
dims_optim = [True] * self.n_dynamic_motor_dims
dims_optim += [True] * (self.n_dynamic_motor_dims * self.n_bfs)
if not optim_end_position:
default[-self.n_dynamic_motor_dims:] = default_motor_end_position or [0.] * self.n_dynamic_motor_dims
dims_optim += [False] * self.n_dynamic_motor_dims
else:
dims_optim += [True] * self.n_dynamic_motor_dims
self.motor_dmp = DmpPrimitive(self.n_dynamic_motor_dims,
self.n_bfs,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
class DmpAgent(Agent):
def __init__(self,
n_dmps,
n_bfs,
used,
default,
conf,
sm,
im,
dmp_type='discrete',
ay=None):
Agent.__init__(self, conf, sm, im)
self.n_dmps, self.n_bfs = n_dmps, n_bfs
self.current_m = zeros(self.conf.m_ndims)
if ay is None:
self.dmp = DmpPrimitive(n_dmps,
n_bfs,
used,
default,
type=dmp_type)
else:
self.dmp = DmpPrimitive(n_dmps,
n_bfs,
used,
default,
type=dmp_type,
ay=ones(n_dmps) * 1.)
@classmethod
def from_settings(cls, n_bfs, starting_position, babbling_name, sm_name,
im_name):
params = get_params(n_bfs, starting_position, babbling_name, sm_name,
im_name)
return cls(**params)
def motor_primitive(self, m):
self.m = bounds_min_max(m, self.conf.m_mins, self.conf.m_maxs)
y = self.dmp.trajectory(self.m)
self.current_m = y[-1, :]
return y # y[:int(len(y) * ((n_bfs*2. - 1.)/(n_bfs*2.))), :]
def sensory_primitive(self, s):
return s[-1] # array([mean(s)]) #s[[-1]]
class DynamicEnvironment(Environment):
def __init__(self, env_cls, env_cfg,
m_mins, m_maxs, s_mins, s_maxs,
n_bfs, move_steps,
n_dynamic_motor_dims, n_dynamic_sensori_dims, max_params,
motor_traj_type="DMP", sensori_traj_type="samples",
optim_initial_position=False, optim_end_position=False, default_motor_initial_position=None, default_motor_end_position=None,
default_sensori_initial_position=None, default_sensori_end_position=None):
self.env = env_cls(**env_cfg)
self.n_bfs = n_bfs
self.n_motor_traj_points = self.n_bfs
self.n_sensori_traj_points = self.n_bfs
self.move_steps = move_steps
self.n_dynamic_motor_dims = n_dynamic_motor_dims
self.n_dynamic_sensori_dims = n_dynamic_sensori_dims
self.max_params = max_params
self.motor_traj_type = motor_traj_type
self.sensori_traj_type = sensori_traj_type
self.optim_initial_position = optim_initial_position
self.optim_end_position = optim_end_position
Environment.__init__(self, m_mins, m_maxs, s_mins, s_maxs)
self.s_traj = None
if self.motor_traj_type == "DMP":
self.init_motor_DMP(optim_initial_position, optim_end_position, default_motor_initial_position, default_motor_end_position)
else:
raise NotImplementedError
if self.sensori_traj_type == "DMP":
self.init_sensori_DMP(optim_initial_position, optim_end_position, default_sensori_initial_position, default_sensori_end_position)
elif self.sensori_traj_type == "samples":
self.samples = np.array(np.linspace(-1, self.move_steps-1, self.n_sensori_traj_points + 1), dtype=int)[1:]
elif self.sensori_traj_type == "end_point":
self.end_point = self.move_steps-1
else:
raise NotImplementedError
def reset(self):
self.env.reset()
def init_motor_DMP(self, optim_initial_position=True, optim_end_position=True, default_motor_initial_position=None, default_motor_end_position=None):
default = np.zeros(self.n_dynamic_motor_dims * (self.n_bfs + 2))
if not optim_initial_position:
default[:self.n_dynamic_motor_dims] = default_motor_initial_position or [0.] * self.n_dynamic_motor_dims
dims_optim = [False] * self.n_dynamic_motor_dims
else:
dims_optim = [True] * self.n_dynamic_motor_dims
dims_optim += [True] * (self.n_dynamic_motor_dims * self.n_bfs)
if not optim_end_position:
default[-self.n_dynamic_motor_dims:] = default_motor_end_position or [0.] * self.n_dynamic_motor_dims
dims_optim += [False] * self.n_dynamic_motor_dims
else:
dims_optim += [True] * self.n_dynamic_motor_dims
self.motor_dmp = DmpPrimitive(self.n_dynamic_motor_dims,
self.n_bfs,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def init_sensori_DMP(self, optim_initial_position=True, optim_end_position=True, default_sensori_initial_position=None, default_sensori_end_position=None):
default = np.zeros(self.n_dynamic_sensori_dims * (self.n_sensori_traj_points + 2))
if not optim_initial_position:
default[:self.n_dynamic_sensori_dims] = default_sensori_initial_position
dims_optim = [False] * self.n_dynamic_sensori_dims
else:
dims_optim = [True] * self.n_dynamic_sensori_dims
dims_optim += [True] * (self.n_dynamic_sensori_dims * self.n_sensori_traj_points)
if not optim_end_position:
default[-self.n_dynamic_sensori_dims:] = default_sensori_end_position
dims_optim += [False] * self.n_dynamic_sensori_dims
else:
dims_optim += [True] * self.n_dynamic_sensori_dims
self.sensori_dmp = DmpPrimitive(self.n_dynamic_sensori_dims,
self.n_sensori_traj_points,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def compute_motor_command(self, m_ag):
m_ag = bounds_min_max(m_ag, self.conf.m_mins, self.conf.m_maxs)
if self.motor_traj_type == "DMP":
dyn_idx = range(self.n_dynamic_motor_dims * self.n_motor_traj_points)
m_weighted = m_ag[dyn_idx] * self.max_params
if self.optim_initial_position:
m_weighted[:self.n_dynamic_motor_dims] = m_weighted[:self.n_dynamic_motor_dims] / self.max_params
if self.optim_end_position:
m_weighted[-self.n_dynamic_motor_dims:] = m_weighted[-self.n_dynamic_motor_dims:] / self.max_params
m_dyn = self.motor_dmp.trajectory(m_weighted)
static_idx = range(self.n_dynamic_motor_dims * self.n_motor_traj_points, self.conf.m_ndims)
m_static = m_ag[static_idx]
m = [list(m_dyn_param) + list(m_static) for m_dyn_param in list(m_dyn)]
else:
raise NotImplementedError
return m
def compute_sensori_effect(self, m_traj):
s = self.env.update(m_traj, reset=False, log=False)
self.s_traj = s
y = np.array(s[:self.move_steps])
if self.sensori_traj_type == "DMP":
self.sensori_dmp.dmp.imitate_path(np.transpose(y))
w = self.sensori_dmp.dmp.w.flatten()
s_ag = list(w)
elif self.sensori_traj_type == "samples":
w = y[self.samples,:]
s_ag = list(np.transpose(w).flatten())
elif self.sensori_traj_type == "end_point":
s_ag = y[self.end_point,:].flatten()
else:
raise NotImplementedError
s = s_ag
return bounds_min_max(s, self.conf.s_mins, self.conf.s_maxs)
def update(self, m_ag, reset=True, log=False):
if reset:
self.reset()
if len(np.array(m_ag).shape) == 1:
s = self.one_update(m_ag, log)
else:
s = []
for m in m_ag:
s.append(self.one_update(m, log))
s = np.array(s)
return s
def plot(self, fig, ax, **kwargs):
ax.cla()
ax.set_aspect('equal')
ax.set_xlim((-1.5, 1.5))
ax.set_ylim((-1.5, 1.5))
def animate(i): return tuple(self.env.plot_update(ax, i))
return animation.FuncAnimation(fig, animate, frames=50, interval=50, blit=True).to_html5_video()
class DivaDMPEnvironment(Environment):
def __init__(self,
synth,
m_mins,
m_maxs,
s_mins,
s_maxs,
m_used,
s_used,
n_dmps,
n_bfs,
dmp_move_steps,
dmp_max_param,
sensory_traj_samples,
audio,
diva_path=None):
self.m_mins = m_mins
self.m_maxs = m_maxs
self.s_mins = s_mins
self.s_maxs = s_maxs
self.m_used = m_used
self.s_used = s_used
self.n_dmps = n_dmps
self.n_bfs = n_bfs
self.dmp_move_steps = dmp_move_steps
self.dmp_max_param = dmp_max_param
self.samples = array(linspace(-1, self.dmp_move_steps - 1,
sensory_traj_samples + 1),
dtype=int)[1:]
self.audio = audio
self.f0 = 1.
self.pressure = 1.
self.voicing = 1.
if self.audio:
import pyaudio
self.pa = pyaudio.PyAudio()
self.stream = self.pa.open(format=pyaudio.paFloat32,
channels=1,
rate=11025,
output=True)
if synth == "octave":
self.synth = DivaOctaveSynth(diva_path)
elif synth == "matlab":
self.synth = DivaMatlabSynth()
else:
raise NotImplementedError
self.art = array(
[0.] * 10 + [self.f0, self.pressure, self.voicing]
) # 13 articulators is a constant from diva_synth.m in the diva source code
self.max_params = array([1.] * self.n_dmps + [self.dmp_max_param] *
self.n_bfs * self.n_dmps + [1.] * self.n_dmps)
self.dmp = DmpPrimitive(dmps=self.n_dmps,
bfs=self.n_bfs,
timesteps=self.dmp_move_steps)
Environment.__init__(self, self.m_mins, self.m_maxs, self.s_mins,
self.s_maxs)
def compute_motor_command(self, m):
return bounds_min_max(m, self.m_mins, self.m_maxs)
def compute_sensori_effect(self, m):
m = self.trajectory(m)
self.art_traj = zeros((13, array(m).shape[0]))
self.art_traj[10, :] = self.f0
self.art_traj[11, :] = self.pressure
self.art_traj[12, :] = self.voicing
self.art_traj[self.m_used, :] = transpose(m)
res = self.synth.execute(2. * (self.art_traj))
formants = log2(transpose(res[self.s_used, :]))
formants[isnan(formants)] = 0.
self.formants_traj = formants
return list(formants[self.samples, 0]) + list(formants[self.samples,
1])
def trajectory(self, m):
y = self.dmp.trajectory(array(m) * self.max_params)
if len(y) > self.dmp_move_steps:
ls = linspace(0, len(y) - 1, self.dmp_move_steps)
ls = array(ls, dtype='int')
y = y[ls]
return y
def update(self, mov, audio=True):
s = Environment.update(self, mov)
if self.audio and audio:
self.play_sound(self.art_traj)
return s
def sound_wave(self, art_traj, power=2.):
synth_art = self.art.reshape(1, -1).repeat(len(art_traj.T), axis=0)
synth_art[:, :] = art_traj.T
return power * self.synth.sound_wave(synth_art.T)
def play_sound(self, art_traj):
sound = self.sound_wave(art_traj)
self.stream.write(sound.astype(float32).tostring())
class DynamicEnvironment(Environment):
def __init__(self,
env_cls,
env_cfg,
m_mins,
m_maxs,
s_mins,
s_maxs,
n_bfs,
move_steps,
n_dynamic_motor_dims,
n_dynamic_sensori_dims,
max_params,
n_sensori_traj_points=None,
motor_traj_type="DMP",
sensori_traj_type="samples",
optim_initial_position=False,
optim_end_position=False,
default_motor_initial_position=None,
default_motor_end_position=None,
default_sensori_initial_position=None,
default_sensori_end_position=None):
self.env = env_cls(**env_cfg)
self.n_bfs = n_bfs
self.n_motor_traj_points = self.n_bfs + optim_initial_position + optim_end_position
self.n_sensori_traj_points = n_sensori_traj_points or self.n_bfs
self.move_steps = move_steps
self.n_dynamic_motor_dims = n_dynamic_motor_dims
self.n_dynamic_sensori_dims = n_dynamic_sensori_dims
self.max_params = max_params
self.motor_traj_type = motor_traj_type
self.sensori_traj_type = sensori_traj_type
self.optim_initial_position = optim_initial_position
self.optim_end_position = optim_end_position
Environment.__init__(self, m_mins, m_maxs, s_mins, s_maxs)
self.s_traj = None
if self.motor_traj_type == "DMP":
self.init_motor_DMP(optim_initial_position, optim_end_position,
default_motor_initial_position,
default_motor_end_position)
else:
raise NotImplementedError
if self.sensori_traj_type == "DMP":
self.init_sensori_DMP(optim_initial_position, optim_end_position,
default_sensori_initial_position,
default_sensori_end_position)
elif self.sensori_traj_type == "samples":
self.samples = np.array(np.linspace(
-1, self.move_steps - 1, self.n_sensori_traj_points + 1),
dtype=int)[1:]
elif self.sensori_traj_type == "end_point":
self.end_point = self.move_steps - 1
else:
raise NotImplementedError
def reset(self):
self.env.reset()
def init_motor_DMP(self,
optim_initial_position=True,
optim_end_position=True,
default_motor_initial_position=None,
default_motor_end_position=None):
default = np.zeros(self.n_dynamic_motor_dims * (self.n_bfs + 2))
if not optim_initial_position:
default[:self.
n_dynamic_motor_dims] = default_motor_initial_position or [
0.
] * self.n_dynamic_motor_dims
dims_optim = [False] * self.n_dynamic_motor_dims
else:
dims_optim = [True] * self.n_dynamic_motor_dims
dims_optim += [True] * (self.n_dynamic_motor_dims * self.n_bfs)
if not optim_end_position:
default[-self.
n_dynamic_motor_dims:] = default_motor_end_position or [
0.
] * self.n_dynamic_motor_dims
dims_optim += [False] * self.n_dynamic_motor_dims
else:
dims_optim += [True] * self.n_dynamic_motor_dims
self.motor_dmp = DmpPrimitive(self.n_dynamic_motor_dims,
self.n_bfs,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def init_sensori_DMP(self,
optim_initial_position=True,
optim_end_position=True,
default_sensori_initial_position=None,
default_sensori_end_position=None):
default = np.zeros(self.n_dynamic_sensori_dims *
(self.n_sensori_traj_points + 2))
if not optim_initial_position:
default[:self.
n_dynamic_sensori_dims] = default_sensori_initial_position
dims_optim = [False] * self.n_dynamic_sensori_dims
else:
dims_optim = [True] * self.n_dynamic_sensori_dims
dims_optim += [True] * (self.n_dynamic_sensori_dims *
self.n_sensori_traj_points)
if not optim_end_position:
default[-self.
n_dynamic_sensori_dims:] = default_sensori_end_position
dims_optim += [False] * self.n_dynamic_sensori_dims
else:
dims_optim += [True] * self.n_dynamic_sensori_dims
self.sensori_dmp = DmpPrimitive(self.n_dynamic_sensori_dims,
self.n_sensori_traj_points,
dims_optim,
default,
type='discrete',
timesteps=self.move_steps)
def compute_motor_command(self, m_ag):
m_ag = bounds_min_max(m_ag, self.conf.m_mins, self.conf.m_maxs)
if self.motor_traj_type == "DMP":
dyn_idx = range(self.n_dynamic_motor_dims *
self.n_motor_traj_points)
m_weighted = m_ag[dyn_idx] * self.max_params
if self.optim_initial_position:
m_weighted[:self.
n_dynamic_motor_dims] = m_weighted[:self.
n_dynamic_motor_dims] / self.max_params
if self.optim_end_position:
m_weighted[-self.n_dynamic_motor_dims:] = m_weighted[
-self.n_dynamic_motor_dims:] / self.max_params
m_dyn = self.motor_dmp.trajectory(m_weighted)
static_idx = range(
self.n_dynamic_motor_dims * self.n_motor_traj_points,
self.conf.m_ndims)
m_static = m_ag[static_idx]
m = [
list(m_dyn_param) + list(m_static)
for m_dyn_param in list(m_dyn)
]
else:
raise NotImplementedError
return m
def compute_sensori_effect(self, m_traj):
s = self.env.update(m_traj, reset=False, log=False)
self.s_traj = s
y = np.array(s[:self.move_steps])
if self.sensori_traj_type == "DMP":
self.sensori_dmp.dmp.imitate_path(np.transpose(y))
w = self.sensori_dmp.dmp.w.flatten()
s_ag = list(w)
elif self.sensori_traj_type == "samples":
w = y[self.samples, :]
s_ag = list(np.transpose(w).flatten())
elif self.sensori_traj_type == "end_point":
s_ag = y[self.end_point, :].flatten()
else:
raise NotImplementedError
s = list(s_ag) + list(s[self.move_steps:])
return bounds_min_max(s, self.conf.s_mins, self.conf.s_maxs)
def update(self, m_ag, reset=True, log=False):
if reset:
self.reset()
if len(np.array(m_ag).shape) == 1:
s = self.one_update(m_ag, log)
else:
s = []
for m in m_ag:
s.append(self.one_update(m, log))
s = np.array(s)
return s
def plot(self, fig, ax, **kwargs):
ax.cla()
ax.set_aspect('equal')
ax.set_xlim((-1.5, 1.5))
ax.set_ylim((-1.5, 1.5))
def animate(i):
return tuple(self.env.plot_update(ax, i))
return animation.FuncAnimation(fig,
animate,
frames=50,
interval=50,
blit=True).to_html5_video()
class DivaDMPEnvironment(Environment):
def __init__(self,
synth,
m_mins, m_maxs, s_mins, s_maxs,
m_used,
s_used,
n_dmps,
n_bfs,
dmp_move_steps,
dmp_max_param,
sensory_traj_samples,
audio,
diva_path=None):
self.m_mins = m_mins
self.m_maxs = m_maxs
self.s_mins = s_mins
self.s_maxs = s_maxs
self.m_used = m_used
self.s_used = s_used
self.n_dmps = n_dmps
self.n_bfs = n_bfs
self.dmp_move_steps = dmp_move_steps
self.dmp_max_param = dmp_max_param
self.samples = array(linspace(-1, self.dmp_move_steps-1, sensory_traj_samples + 1), dtype=int)[1:]
self.audio = audio
self.f0 = 1.
self.pressure = 1.
self.voicing = 1.
if self.audio:
import pyaudio
self.pa = pyaudio.PyAudio()
self.stream = self.pa.open(format=pyaudio.paFloat32,
channels=1,
rate=11025,
output=True)
if synth == "octave":
self.synth = DivaOctaveSynth(diva_path)
elif synth == "matlab":
self.synth = DivaMatlabSynth()
else:
raise NotImplementedError
self.art = array([0.]*10 + [self.f0, self.pressure, self.voicing]) # 13 articulators is a constant from diva_synth.m in the diva source code
self.max_params = array([1.] * self.n_dmps + [self.dmp_max_param] * self.n_bfs * self.n_dmps + [1.] * self.n_dmps)
self.dmp = DmpPrimitive(dmps=self.n_dmps, bfs=self.n_bfs, timesteps=self.dmp_move_steps)
Environment.__init__(self, self.m_mins, self.m_maxs, self.s_mins, self.s_maxs)
def compute_motor_command(self, m):
return bounds_min_max(m, self.m_mins, self.m_maxs)
def compute_sensori_effect(self, m):
m = self.trajectory(m)
self.art_traj = zeros((13, array(m).shape[0]))
self.art_traj[10, :] = self.f0
self.art_traj[11, :] = self.pressure
self.art_traj[12, :] = self.voicing
self.art_traj[self.m_used,:] = transpose(m)
res = self.synth.execute(2.*(self.art_traj))
formants = log2(transpose(res[self.s_used,:]))
formants[isnan(formants)] = 0.
self.formants_traj = formants
return list(formants[self.samples,0]) + list(formants[self.samples,1])
def trajectory(self, m):
y = self.dmp.trajectory(array(m) * self.max_params)
if len(y) > self.dmp_move_steps:
ls = linspace(0,len(y)-1,self.dmp_move_steps)
ls = array(ls, dtype='int')
y = y[ls]
return y
def update(self, mov, audio=True):
s = Environment.update(self, mov)
if self.audio and audio:
self.play_sound(self.art_traj)
return s
def sound_wave(self, art_traj, power=2.):
synth_art = self.art.reshape(1, -1).repeat(len(art_traj.T), axis=0)
synth_art[:, :] = art_traj.T
return power * self.synth.sound_wave(synth_art.T)
def play_sound(self, art_traj):
sound = self.sound_wave(art_traj)
self.stream.write(sound.astype(float32).tostring())
