def predict_dialogue(self, user_utt, sys_utt, da, user_start_time):
s_len = len(sys_utt)
u_len = len(user_utt)
diff = s_len - u_len
df = pf.makeDiaDF(user_start_time, s_len, u_len, diff, da)
X_test = df.iloc[0, :].values
d_pred = pf.predUnknown(X_test, "dialogue")
return d_pred
def predict_text(self, user_utt):
df = etf.makeFea(user_utt)
pca = pickle.load(
open("../refData/RL_files/modelUIpred/pca.model", "rb"))
df = pf.PCAonlyBOW(df, pca=pca)
X_test = df.iloc[0, 1:-1].values
X_test = X_test.astype(np.float32)
t_pred = pf.predUnknown(X_test, "text")
return t_pred
# 心象を線形変換します
def changePredValueRange(fusion_pred):
lowUI_before = x1 = 4.4
lowUI_after = y1 = 3
highUI_before = x2 = 4.6
highUI_after = y2 = 5
fusion_pred_scaled = ((y2 - y1) / (x2 - x1)) * (fusion_pred - x1) + y1
if fusion_pred_scaled < 1:
fusion_pred_scaled = 1
if fusion_pred_scaled > 7:
fusion_pred_scaled = 7
return fusion_pred_scaled
if __name__ == '__main__':
# ここは関数の保持のみを行うコード
print('it is the code has main function.')
#### 推定値のrangeの調整
import predUI as pf
d_pred, v_pred, t_pred, f_pred = 4.0, 4.0, 0.1, 0.1
df = pd.DataFrame(data=[[d_pred, v_pred, t_pred, f_pred]])
X_test = df.iloc[0, :].values
user_impression = pf.predUnknown(X_test, 'fusion')
def predict_face(self, start, end):
file_name = "{}/{}_face.csv".format(self.data_path, self.user)
X_test = eff.predictionFace(start, end, file_name)
f_pred = pf.predUnknown(X_test, "face")
return f_pred
def predict_voice(self, file_name):
df = pd.read_csv(file_name)
df = pf.selectVoiceFea(df, case="pred")
X_test = df.iloc[0, :].values
v_pred = pf.predUnknown(X_test, "voice")
return v_pred
def process(self):
chg_theme, theme = self.themeHis.decideNextTheme(self.user_impression)
# システム発話生成
sys_utt, da = self.env.utterance_selection_softmax(\
chg_theme, theme, self.agent.Q[self.state])
# 発話選択
self.env.sys_utterance_log.append(sys_utt)
file_name = "../../user_data/{0}/voice/{0}_{1}_voice"\
.format(self.user, str(self.current_turn).zfill(3))
self.record_log(sys_utt)
print("system:{}".format(sys_utt))
sys_start_time = time.time() - self.base_time
self.mmd.say(sys_utt)
self.record_log("*")
self.record.start(file_name)
self.asr.start()
while True:
if self.asr.m_turn.is_sys_turn_end == True:
break
else:
time.sleep(0.1)
self.record.stop()
user_turn_end_time = time.time() - self.base_time
user_utt = self.asr.read_result()
self.env.user_utterance_log.append(user_utt)
user_utt_start_time = self.asr.get_utt_start_time()
# 録音したwavファイルが保存されるまで待機
while True:
if self.record.save_complete == True:
time.sleep(0.1)
break
else:
time.sleep(0.1)
self.opensmile.run(file_name)
print("user:{}".format(user_utt))
# predict mono modal
d_pred = self.predict_dialogue(user_utt, sys_utt, da,
user_utt_start_time)
v_pred = self.predict_voice("{}.csv".format(file_name))
t_pred = self.predict_text(user_utt)
f_pred = self.predict_face(sys_start_time, user_turn_end_time)
self.print_debug("dialogue predict:{}".format(d_pred))
self.print_debug("voice predict:{}".format(v_pred))
self.print_debug("text predict:{}".format(t_pred))
self.print_debug("face predict:{}".format(f_pred))
# fusion 4 modal
df = pd.DataFrame(data=[[d_pred, v_pred, t_pred, f_pred]])
X_test = df.iloc[0, :].values
self.user_impression = pf.predUnknown(X_test, "fusion")
# 値のrange変更
self.user_impression = pf.changePredValueRange(self.user_impression)
self.print_debug("user_impression:{}".format(self.user_impression))
# 状態更新
n_state = self.env.get_next_state(self.user_impression, sys_utt,
user_utt)
self.state = n_state
# 終了判定
self.current_turn += 1
if self.current_turn == self.turn_num:
is_end = True
else:
is_end = False
return is_end