def trigger_user_turn_off(self):
should_respond = False
if self.cns.user_turn_active:
if not self.cns.vad_ipu_active and not self.cns.asr_active:
current_utt = clean_whitespace(self.cns.user.utterance)
if current_utt != self.last_current_utterance:
self.last_current_utterance = current_utt
context = self.cns.memory.get_dialog_text()
context.append(current_utt)
trp = self.lm_eot(context)
self.cns.user.all_trps.append({
"trp": trp,
"time": time.time()
})
self.last_guess = "trp"
if trp >= self.trp_threshold:
should_respond = True
self.cns.user.trp_at_eot = trp
self.cns.user.utterance_at_eot = current_utt
self.cns.finalize_user()
if self.verbose:
print(C.green +
f"EOT recognized: {round(trp, 3)}" + C.end)
else:
if self.verbose:
if self.last_guess != "listen":
print(C.red + f"listen: {1-round(trp, 3)}" +
C.end)
self.last_guess = "listen"
return should_respond
def get_dialog_text_debug(self):
turns = self.get_turns()
dialog = []
if len(turns) > 0:
utt = str(turns[0].start_time) + turns[0].utterance
last_name = turns[0].name
for t in turns[1:]:
if t.name == last_name:
utt = utt + " " + str(t.start_time) + t.utterance
else:
dialog.append(clean_whitespace(utt))
utt = str(t.start_time) + t.utterance
last_name = t.name
dialog.append(clean_whitespace(utt))
return dialog
def get_dialog_text(self):
turns = self.get_turns()
# utt = str(turns[0].start_time) + turns[0].utterance
dialog = []
if len(turns) > 0:
utt = turns[0].utterance
last_name = turns[0].name
for t in turns[1:]:
if t.utterance == "":
continue
if t.name == last_name:
utt = utt + " " + clean_whitespace(t.utterance)
else:
dialog.append(clean_whitespace(utt))
# utt = str(t.start_time) + t.utterance
utt = clean_whitespace(t.utterance)
last_name = t.name
dialog.append(clean_whitespace(utt))
return dialog, last_name
def get_dialog_text(all_turns, turn_index):
turns = all_turns[:turn_index + 1]
# utt = str(turns[0].start_time) + turns[0].utterance
dialog = []
if len(turns) > 0:
utt = turns[0]["utterance"]
last_name = turns[0]["name"]
for t in turns[1:]:
if t["utterance"] == "":
continue
if t["name"] == last_name:
utt = utt + " " + clean_whitespace(t["utterance"])
else:
dialog.append(clean_whitespace(utt))
# utt = str(t.start_time) + t['utterance']
utt = clean_whitespace(t["utterance"])
last_name = t["name"]
dialog.append(clean_whitespace(utt))
return dialog
def trigger_user_turn_off(self):
should_respond = False
if self.cns.user_turn_active and not self.cns.vad_ipu_active:
current_utt = clean_whitespace(self.cns.user.prel_utterance)
if current_utt != "" and current_utt != self.last_current_utterance:
# print("guessing")
# update last prediction text
self.last_current_utterance = current_utt
# get result from prediction model
result = self.predict_trp(current_utt)
trp = result["p"]
# append trp estimate, the current time and duration of prediction
print("pred time: ", result["time"])
self.cns.user.all_trps.append({
"trp":
trp,
"utterance":
current_utt,
"predictions":
result["predictions"],
"prediction_time":
result["time"],
"time":
time.time(),
})
self.cns.user.pred_time.append(result["time"])
# self.last_guess = "trp"
# check if a turn-shift is recognized
if trp >= self.trp_threshold:
print("pos trp")
self.cns.user.trp_at_eot = trp
self.cns.user.utterance_at_eot = current_utt
# self.t_predicted_user_off = time.time()
# print("last_agent_trigger_on: ", self.last_agent_trigger_on)
# print("should respond: TRUE")
should_respond = True
# self.cns.finalize_user()
if self.verbose:
print(C.green + f"Take turn: {round(trp, 3)}" + C.end)
else:
print("neg trp")
if self.verbose:
print(C.green + f"Listen: ", C.red,
f"{round(trp, 3)}" + C.end)
# self.last_guess = "listen"
return should_respond
def finalize(self):
super().finalize()
self.planned_utterance = clean_whitespace(self.planned_utterance)
if self.completion >= 1:
self.utterance = self.planned_utterance
def finalize(self):
self.utterance = clean_whitespace(self.utterance)
self.end_time = time.time()