def rasa_nlu():
m1 = MicrophoneModule(5000)
m2 = GoogleASRModule("en-US")
m3 = CallbackModule(callback=lambda x: print("%s (%f) - %s" % (
x.text, x.confidence, x.final)))
m4 = RasaNLUModule("data/rasa/models/nlu/default/current")
m5 = CallbackModule(callback=lambda x: print(x.act, "-", x.concepts))
m1.subscribe(m2)
m2.subscribe(m3)
m2.subscribe(m4)
m4.subscribe(m5)
m4.setup()
m2.setup()
m1.setup()
m3.setup()
m5.setup()
print("All setup")
m1.run(run_setup=False)
m2.run(run_setup=False)
m3.run(run_setup=False)
m4.run(run_setup=False)
m5.run(run_setup=False)
input()
m1.stop()
m2.stop()
m3.stop()
m4.stop()
m5.stop()
def repeat_demo():
m1 = MicrophoneModule(5000)
m2 = GoogleASRModule("en-US")
m3 = TextDispatcherModule()
m4 = GoogleTTSModule("en-US", "en-US-Wavenet-A")
m5 = AudioDispatcherModule(5000)
m6 = StreamingSpeakerModule(5000)
m1.subscribe(m2)
m2.subscribe(m3)
m3.subscribe(m4)
m4.subscribe(m5)
m5.subscribe(m6)
m1.setup()
m2.setup()
m3.setup()
m4.setup()
m5.setup()
m6.setup()
print("All setup")
m1.run(run_setup=False)
m2.run(run_setup=False)
m3.run(run_setup=False)
m4.run(run_setup=False)
m5.run(run_setup=False)
m6.run(run_setup=False)
input()
m1.stop()
m2.stop()
m3.stop()
m4.stop()
m5.stop()
m6.stop()
def google_asr():
m1 = MicrophoneModule(5000)
m2 = GoogleASRModule("en-US") # en-US or de-DE or ....
m3 = CallbackModule(callback=lambda x: print("%s (%f) - %s" % (
x.text, x.stability, x.final)))
m1.subscribe(m2)
m2.subscribe(m3)
m1.run()
m2.run()
m3.run()
input()
m1.stop()
m2.stop()
m3.stop()
def audio():
rate = 16000
chunk_time = 0.1
chunk_size = int(rate * chunk_time)
sample_width = 2
m1 = MicrophoneModule(chunk_size, rate=rate)
m2 = StreamingSpeakerModule(chunk_size, rate)
# db = DebugModule()
db = CustomModule(chunk_size, rate=rate)
m1.subscribe(m2)
m1.subscribe(db)
m1.run()
m2.run()
db.run()
input()
m1.stop()
m2.stop()
db.stop()
from retico.modules.azure.emotion_recognition import AzureEmotionDetectionModule
from retico.interop.ros.catkin_ws.src.rosretico.src.RosPublisher import RosPublisher
from retico.interop.ros.catkin_ws.src.rosretico.src.RosSubscriber import RosSubscriber
# how to restart an utterance?
# had to change is_running to _is_running because opendial modules have an is_running() method
# run: export GOOGLE_APPLICATION_CREDENTIALS=/home/casey/substutute-ca5bdacf1d9a.json
aod_endpoint = "https://slimcomputervision.cognitiveservices.azure.com/"
aod_key = "59bfd2dc248a4d08957edf7a6eb6331f"
aed_endpoint = "https://westus2.api.cognitive.microsoft.com/"
aed_key = "1837b9d29e0b4a22843d103a7ca8b3c9"
# instantiate modules
mic = MicrophoneModule(1000, rate=16000)
asr = AzureASRModule("179eaa4b8fc54e0fa5115ba5d14883f2")
emotion = AzureEmotionDetectionModule(aed_key, aed_endpoint)
iasr = IncrementalizeASRModule()
#initialize ros and its topics
rospy.init_node("Robot_Ready_SDS_RosNode", anonymous=True)
mic_publisher = RosPublisher('mic')
asr_publisher = RosPublisher('asr')
# asr_subscriber = RosSubscriber('asr',debug=True)
debug = DebugModule()
# hook modules up to each other
mic.subscribe(asr)
mic.subscribe(mic_publisher)
opendial_variables = ['face_count', #
'num_objs', # ObjectDetector
'exploring', # MistyReferModule
'aligned', # MistyReferModule
'word_to_find', # WordsAsClassifiersModule
'best_object', # WordsAsClassifiersModule
'obj_confidence'] # WordsAsClassifiersModule
#
# INSTANTIATE MODULES
#
# mic = RespeakerMicrophoneModule('192.168.20.49:8000')
# asr = GoogleASRModule(rate=16000)
mic = MicrophoneModule(1000)
asr = GoogleASRModule()
iasr = IncrementalizeASRModule()
dm = OpenDialModule(domain_dir=domain_dir, variables=opendial_variables)
# cozmo_camera = WebcamModule()
misty_camera = MistyCameraModule(misty_ip)
misty_refer = MistyReferModule(misty_ip)
misty_state = MistyStateModule(misty_ip)
cropper = ImageCropperModule(top=200)
# object_detector = AzureObjectDetectionModule(aod_key, aod_endpoint)
object_detector = MaskrRCNNObjectDetection(mask_rcnn_labels, mask_rcnn_model, max_objs=1)
feature_extractor = KerasObjectFeatureExtractorModule()
wac = WordsAsClassifiersModule(wac_dir=wac_dir)
debug = DebugModule()
print('All modules instantiated.')
def __init__(
self,
chunk_time=None,
chunk_size=None,
sample_rate=16000,
bytes_per_sample=2,
language="en-US",
nchunks=20,
use_asr=True,
use_iasr=False,
record=False,
debug=False,
bypass=False,
cache_dir="/tmp",
):
self.sample_rate = sample_rate
assert (chunk_time is not None or chunk_size
is not None), "please provide either chunk_time or chunk_size"
if chunk_size is not None:
self.chunk_size = int(chunk_size)
self.chunk_time = chunk_size / sample_rate
else:
self.chunk_time = chunk_time
self.chunk_size = int(chunk_time * sample_rate)
self.bytes_per_sample = bytes_per_sample
self.use_asr = use_asr
self.use_iasr = use_iasr
self.record = record
self.debug = debug
self.cache_dir = cache_dir
if self.use_iasr:
self.use_asr = True
# Components that are always used
if bypass:
self.in_mic = MicrophoneOutputBypass(
chunk_size=self.chunk_size,
rate=self.sample_rate,
sample_width=self.bytes_per_sample,
)
else:
self.in_mic = MicrophoneModule(
chunk_size=self.chunk_size,
rate=self.sample_rate,
sample_width=self.bytes_per_sample,
)
self.vad_frames = VADFrames(
chunk_time=self.chunk_time,
sample_rate=self.sample_rate,
mode=3,
debug=debug,
)
self.in_mic.subscribe(self.vad_frames)
# Optional Components
if self.use_asr:
self.asr = GoogleASRModule(
language=language,
nchunks=nchunks, # m chunks to trigger a new prediction
rate=self.sample_rate,
)
self.in_mic.subscribe(self.asr)
if self.use_iasr:
self.iasr = IncrementalizeASRModule(
threshold=0.8
) # Gets only the newly added words at each increment
self.asr.subscribe(self.iasr)
if self.record:
self.cache_dir = join(cache_dir, "hearing")
makedirs(self.cache_dir, exist_ok=True)
print("Hearing: ", self.cache_dir)
wav_filename = join(self.cache_dir, "user_audio.wav")
self.audio_record = AudioRecorderModule(
wav_filename, rate=sample_rate, sample_width=bytes_per_sample)
self.in_mic.subscribe(self.audio_record)
if self.debug:
self.asr_debug = ASRDebugModule()
if self.use_asr:
self.asr.subscribe(self.asr_debug)
logging.info(f"{self.name}: Initialized @ {time.time()}")
class Hearing(object):
"""
The Hearing component of the Spoken Dialog System.
Components:
- MicrophoneModule
- ASRModule
- IncrementalizeASRModule
"""
CACHE_DIR = "/tmp"
def __init__(
self,
chunk_time=None,
chunk_size=None,
sample_rate=16000,
bytes_per_sample=2,
language="en-US",
nchunks=20,
use_asr=True,
use_iasr=False,
record=False,
debug=False,
bypass=False,
cache_dir="/tmp",
):
self.sample_rate = sample_rate
assert (chunk_time is not None or chunk_size
is not None), "please provide either chunk_time or chunk_size"
if chunk_size is not None:
self.chunk_size = int(chunk_size)
self.chunk_time = chunk_size / sample_rate
else:
self.chunk_time = chunk_time
self.chunk_size = int(chunk_time * sample_rate)
self.bytes_per_sample = bytes_per_sample
self.use_asr = use_asr
self.use_iasr = use_iasr
self.record = record
self.debug = debug
self.cache_dir = cache_dir
if self.use_iasr:
self.use_asr = True
# Components that are always used
if bypass:
self.in_mic = MicrophoneOutputBypass(
chunk_size=self.chunk_size,
rate=self.sample_rate,
sample_width=self.bytes_per_sample,
)
else:
self.in_mic = MicrophoneModule(
chunk_size=self.chunk_size,
rate=self.sample_rate,
sample_width=self.bytes_per_sample,
)
self.vad_frames = VADFrames(
chunk_time=self.chunk_time,
sample_rate=self.sample_rate,
mode=3,
debug=debug,
)
self.in_mic.subscribe(self.vad_frames)
# Optional Components
if self.use_asr:
self.asr = GoogleASRModule(
language=language,
nchunks=nchunks, # m chunks to trigger a new prediction
rate=self.sample_rate,
)
self.in_mic.subscribe(self.asr)
if self.use_iasr:
self.iasr = IncrementalizeASRModule(
threshold=0.8
) # Gets only the newly added words at each increment
self.asr.subscribe(self.iasr)
if self.record:
self.cache_dir = join(cache_dir, "hearing")
makedirs(self.cache_dir, exist_ok=True)
print("Hearing: ", self.cache_dir)
wav_filename = join(self.cache_dir, "user_audio.wav")
self.audio_record = AudioRecorderModule(
wav_filename, rate=sample_rate, sample_width=bytes_per_sample)
self.in_mic.subscribe(self.audio_record)
if self.debug:
self.asr_debug = ASRDebugModule()
if self.use_asr:
self.asr.subscribe(self.asr_debug)
logging.info(f"{self.name}: Initialized @ {time.time()}")
@property
def name(self):
return self.__class__.__name__
def __repr__(self):
s = "\n" + "=" * 40
s += "\n" + self.__class__.__name__
s += f"\nsample_rate: {self.sample_rate}"
s += f"\nchunk_time: {self.chunk_time}"
s += f"\nchunk_size: {self.chunk_size}"
s += f"\nbytes_per_sample: {self.bytes_per_sample}"
s += f"\nrecord: {self.record}"
s += f"\nuse_asr: {self.use_asr}"
s += f"\ndebug: {self.debug}"
s += "\n" + "=" * 40
return s
def setup(self, **kwargs):
self.in_mic.setup(**kwargs)
if self.use_asr:
self.asr.setup(**kwargs)
if self.use_iasr:
self.iasr.setup(**kwargs)
if self.record:
self.audio_record.setup(**kwargs)
logging.info(f"{self.name}: Setup")
def run(self, **kwargs):
self.in_mic.run(**kwargs)
self.vad_frames.run(**kwargs)
if self.use_asr:
self.asr.run(**kwargs)
if self.use_iasr:
self.iasr.run(**kwargs)
if self.record:
self.audio_record.run(**kwargs)
if self.debug:
if self.use_asr:
self.asr_debug.run(**kwargs)
# self.iasr_debug.run(run_setup=run_setup)
logging.info(f"{self.name}: run @ {time.time()}")
def stop(self, **kwargs):
self.in_mic.stop(**kwargs)
self.vad_frames.stop(**kwargs)
if self.use_asr:
self.asr.stop(**kwargs)
if self.use_iasr:
self.iasr.stop(**kwargs)
if self.record:
self.audio_record.stop(**kwargs)
if self.debug:
if self.use_asr:
self.asr_debug.stop(**kwargs)
logging.info(f"{self.name}: stop_components @ {time.time()}")
def init_all(robot : cozmo.robot.Robot):
domain_dir = '/home/casey/git/retico/data/cozmo/dm/dialogue.xml'
aod_endpoint = ""
aod_key = ""
wac_dir = '/home/casey/git/retico/data/wac/subset'
mask_rcnn_labels = '/home/casey/git/retico/data/maskrcnn/label_map.pbtxt'
mask_rcnn_model = '/home/casey/git/retico/data/maskrcnn/frozen_inference_graph.pb'
opendial_variables = ['face_count', # CozmoStateModule
'num_objs', # ObjectDetector
'near_object', # CozmoRefer
'exploring', # CozmoRefer
'aligned', # CozmoRefer
'word_to_find', # WordsAsClassifiersModule
'best_object', # WordsAsClassifiersModule
'obj_confidence'] # WordsAsClassifiersModule
#
# INSTANTIATE MODULES
#
# mic = RespeakerMicrophoneModule('192.168.0.102:8000')
# asr = GoogleASRModule(rate=16000)
mic = MicrophoneModule(1000)
asr = GoogleASRModule()
iasr = IncrementalizeASRModule()
dm = OpenDialModule(domain_dir=domain_dir, variables=opendial_variables)
cozmo_refer = CozmoReferModule(robot)
cozmo_camera = CozmoCameraModule(robot)
# cozmo_camera = WebcamModule()
# cozmo_state = CozmoStateModule(robot)
# object_detector = AzureObjectDetectionModule(aod_key, aod_endpoint)
object_detector = MaskrRCNNObjectDetection(mask_rcnn_labels, mask_rcnn_model)
feature_extractor = KerasObjectFeatureExtractorModule()
wac = WordsAsClassifiersModule(wac_dir=wac_dir)
debug = DebugModule()
# psi related modules
# WriterSingleton should use the *source* ip address (i.e., this machine's)
# WriterSingleton(ip='192.168.0.101', port='12346') # create the zeromq writer
# psi = ZeroMQWriter(topic='retico')
# mic as input
mic.subscribe(asr)
asr.subscribe(iasr)
iasr.subscribe(wac)
wac.subscribe(dm)
dm.subscribe(cozmo_refer)
# robot state as input
# cozmo_state.subscribe(dm)
wac.subscribe(cozmo_refer)
object_detector.subscribe(dm)
object_detector.subscribe(cozmo_refer)
cozmo_refer.subscribe(dm)
# robot camera as input
cozmo_camera.subscribe(object_detector)
object_detector.subscribe(feature_extractor)
feature_extractor.subscribe(wac)
# feature_extractor.subscribe(debug)
# iasr.subscribe(psi)
# wac.subscribe(psi)
# dm.subscribe(psi)
# feature_extractor.subscribe(psi)
# object_detector.subscribe(psi)
#
# INITIALIZE MODULES
#
mic.run()
asr.run()
iasr.run()
dm.run()
cozmo_refer.run() # IF I MAKE THIS RUN EVERYTHING SLOWS DOWN
# cozmo_state.run()
cozmo_camera.run()
object_detector.run()
feature_extractor.run()
wac.run()
debug.run()
# psi.run()
input() # keep everything running
mic.stop()
asr.stop()
iasr.stop()
cozmo_refer.stop()
# cozmo_state.stop()
dm.stop()
cozmo_camera.stop()
object_detector.stop()
feature_extractor.stop()
wac.stop()
debug.stop()
def audio_demo():
m1 = MicrophoneModule(5000)
m2 = StreamingSpeakerModule(5000)
m1.subscribe(m2)
m1.run()
m2.run()
input()
m1.stop()
m2.stop()
input()
m1.run()
m2.run()
input()
m1.stop()
m2.stop()
m1.run()
m2.run()
db.run()
input()
m1.stop()
m2.stop()
db.stop()
if __name__ == "__main__":
# google_asr()
m1 = MicrophoneModule(5000)
m2 = GoogleASRModule("en-US") # en-US or de-DE or ....
m3 = CustomTextModule()
m1.subscribe(m2)
m2.subscribe(m3)
m1.run()
m2.run()
m3.run()
input()
m1.stop()
m2.stop()
m3.stop()