six-class svc model, increased sensing frequency, semilogy spectrum

A_record.py

@@ -35,9 +35,9 @@ from glob import glob
 # ==================
 BASE_DIR = "."
 SOUND_NAME = "sweep"  # sound to use
-CLASS_LABELS = ["tip", "middle", "base", "back", "none"]  # classes to train
+CLASS_LABELS = ["tip", "middle", "base", "back", "left", "right", "none"]  # classes to train
 SAMPLES_PER_CLASS = 20
-MODEL_NAME = "lndw2022_sweep_1s"
+MODEL_NAME = "berlinsummit_sweep_1s"
 SHUFFLE_RECORDING_ORDER = False
 APPEND_TO_EXISTING_FILES = True
 # ==================
@@ -83,9 +83,10 @@ def setup_experiment():
     current_idx = 0
 
     if APPEND_TO_EXISTING_FILES:
-        max_id = max([int(x.split("/")[-1].split("_")[0]) for x in glob(DATA_DIR+"/*.wav")])
-        label_list = [""]*max_id + label_list
-        current_idx = max_id
+        if len(glob(DATA_DIR+"/*.wav")) > 0:
+            max_id = max([int(x.split("/")[-1].split("_")[0]) for x in glob(DATA_DIR+"/*.wav")])
+            label_list = [""]*max_id + label_list
+            current_idx = max_id
 
 
 def setup_jack(sound_name):

B_train.py

@@ -25,6 +25,9 @@ import pickle
 from sklearn.neighbors import KNeighborsClassifier, KNeighborsRegressor
 from sklearn.model_selection import train_test_split
 from matplotlib import pyplot
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import StandardScaler
+from sklearn.svm import SVC
 
 from A_record import MODEL_NAME
 
@@ -32,7 +35,7 @@ from A_record import MODEL_NAME
 # USER SETTINGS
 # ==================
 BASE_DIR = "."
-SENSORMODEL_FILENAME = "sensor_model_lndw2022_AAS016.pkl"
+SENSORMODEL_FILENAME = "sensor_model_berlinsummit_AAS016_svc.pkl"
 TEST_SIZE = 0  # percentage of samples left out of training and used for reporting test score
 SHOW_PLOTS = True
 # ==================
@@ -138,7 +141,7 @@ def main():
     sounds, labels = load_sounds(DATA_DIR)
     # spectra = [sound_to_spectrum(sound) for sound in sounds]
     spectra = [sound_to_spectrum_stft(sound) for sound in sounds]
-    classes = list(set(labels))
+    classes = sorted(list(set(labels)))
 
     if SHOW_PLOTS:
         plot_spectra(spectra, labels)
@@ -148,8 +151,15 @@ def main():
     else:
         X_train, y_train = (spectra, labels)
 
-    clf = KNeighborsClassifier()  # using default KNN classifier
+    # clf = KNeighborsClassifier()  # using default KNN classifier
+    clf = Pipeline([
+        ("standardscaler", StandardScaler()),
+        ("classifier", SVC(kernel="linear", C=0.01,
+                           probability=True)),
+    ])
+    clf.name = "svc_custom"  # name is used for saving
     clf.fit(X_train, y_train)
+
     train_score = clf.score(X_train, y_train)
     print("Fitted sensor model to data!")
     print("Training score: {:.2f}".format(train_score))

C_sense.py

@@ -49,7 +49,17 @@ label_renamer_lndw2022 = {
     "back": "Rückseite",
     "none": "Ohne"
 }
-label_renamer = label_renamer_lndw2022
+label_renamer_berlinsummit = {
+    "tip": "Tip",
+    "middle": "Middle",
+    "base": "Base",
+    "back": "Backside",
+    "none": "No contact",
+    "left": "Left side",
+    "right": "Right side"
+}
+
+label_renamer = label_renamer_berlinsummit
 
 CHANNELS = 1
 SR = 48000
@@ -72,6 +82,7 @@ class LiveAcousticSensor(object):
         self.setup_jack()
         self.setup_model()
         self.setup_window()
+        self.pause_time = 0.01
 
     def setup_jack(self):
         self.J = JackSignal("JS")
@@ -115,11 +126,13 @@ class LiveAcousticSensor(object):
         ax2.set_title("Amplitude spectrum", size=20)
         ax2.set_xlabel("Frequency [Hz]")
         self.wavelines, = ax1.plot(self.Ains[0])
-        self.spectrumlines, = ax2.plot(sound_to_spectrum_stft(self.Ains[0]))
-        ax2.set_ylim([0, 250])
+        self.spectrumlines, = ax2.plot(sound_to_spectrum_stft(self.Ains[0])+1e-16)
+        ax2.set_ylim([1e-1, 1000])
+        ax2.set_yscale("log")
+
 
         ax3a = f.add_subplot(2, 2, 3)
-        ax3a.text(0.0, 0.8, "Gemessener Kontakt:", dict(size=40))
+        ax3a.text(0.0, 0.8, "Measured contact:", dict(size=40))
         self.predictiontext = ax3a.text(0.25, 0.25, "", dict(size=70))
         ax3a.set_xticklabels([])
         ax3a.set_yticklabels([])
@@ -169,7 +182,7 @@ class LiveAcousticSensor(object):
                     self.J.process()
                     self.J.wait()
                     self.predict()
-                plt.pause(1)
+                plt.pause(self.pause_time)
         else:
             key = input("Press <Enter> to sense! ('q' to abort)")
             while key == '':
@@ -189,6 +202,12 @@ def on_key(event):
         toggle_pause(event)
     elif event.key == "q":
         sys.exit()
+    elif event.key == "+":
+        predictor.pause_time = min(predictor.pause_time + 0.2, 2)
+        print(f"{predictor.pause_time=}")
+    elif event.key == "-":
+        predictor.pause_time = max(predictor.pause_time - 0.2, 0)
+        print(f"{predictor.pause_time=}")
 
 
 def main():

set_volume.sh

+#!/usr/bin/env bash
+amixer -D hw:USB sset 'Master',0 80% unmute
+amixer -D hw:USB sset 'Line',0 0% mute cap capture 70%
+amixer -D hw:USB sset 'Line',1 0% mute