def update_equiv_src(self, new_src: Source):
""""If the new source is within the margin of an old source, add its audio to the old source and
analyze with ML"""
logging.debug(f"Updating equivalent source...")
for src in self.sources:
if src.position.within_margin(resonant.SOURCE_MARGIN,
new_src.position):
logging.debug(
f"New source ({new_src}) within margin of old source ({src})... updating"
)
src.update_audio(new_src.audio[:resonant.LOCALIZING_WINDOW])
src.position = new_src.position
# It only runs ml if it has enough samples to analyze
"""if src.can_ml_analyze:
# self.ml.analyze(src)
# This means the ML was conclusive and it keeps it alive
if src.name is not None:
src.reset_cycles()
"""
return
# If there are no equivalent sources, add it to the tracked sources
new_src.track()
self.sources.append(new_src)
logging.debug(
f"No matching source to new source ({new_src}). Adding to tracked sources."
)
def test_src_filter(sources: List[Source]):
resonant.CYCLES_TO_LIVE = 3
sources = [
Source(60, np.array([1, 1, 1, 1, 1, 1, 1])),
Source(21, np.array([1, 1, 1, 1, 1, 1, 1])),
]
sources[0].cycles_lived = 2
sources[1].cycles_lived = 0
def test_ml_max_samples(self):
resonant.MIN_ML_SAMPLES = 6
resonant.MAX_ML_SAMPLES = 10
src = Source(33, np.zeros(7))
assert src.can_ml_analyze is True
src.audio = np.zeros(11)
with pytest.raises(AssertionError) as exc_info:
src.can_ml_analyze
def test_ml_min_samples(self):
resonant.MIN_ML_SAMPLES = 6
resonant.MAX_ML_SAMPLES = 10
src = Source(33, np.zeros(5))
assert src.can_ml_analyze is False
src.audio = np.zeros(6)
assert src.can_ml_analyze is True
def sources():
empty_audio = np.zeros(5)
resonant.SOURCE_MARGIN = SphericalPt.angle_only(5)
resonant.MAX_ML_SAMPLES = 8
resonant.MIN_ML_SAMPLES = 6
sources = [
Source(30, empty_audio),
Source(88, empty_audio),
]
return sources
def test_updates(count_nan):
resonant.MIN_ML_SAMPLES = 3
resonant.MAX_ML_SAMPLES = 8
src = Source(30, np.array([]))
assert src.can_ml_analyze is False
assert count_nan(src.audio) == resonant.MAX_ML_SAMPLES
additional = np.array([2, 2, 2, 2])
src.update_audio(additional)
expected = np.array([2, 2, 2, 2, np.nan, np.nan, np.nan, np.nan])
assert np.array_equal(src.audio, expected, equal_nan=True) is True
assert src.can_ml_analyze is True
assert count_nan(src.audio) == 4
def run_algorithm(cls, microphones: List[Mic]):
def get_ratio(m1, m2):
index_delay = fft_crosscorr(
m1.signal, m2.signal).argmax() - len(m1.signal) / 2
confidence = np.real(fft_crosscorr(m1.signal, m2.signal).max())
ratio = resonant.V_SOUND * index_delay / \
(resonant.AUDIO_SAMPLING_RATE * resonant.MIC_SPACING * math.sqrt(2))
ratio = np.clip(ratio, -1, 1)
return ratio, confidence
def ave_angle(a1, a2):
if (a1 < 90 and a2 > 270):
a1 -= 360
if (a2 < 90 and a1 > 270):
a2 -= 360
ave = (a1 + a2) / 2
if (ave < 0):
ave += 360
return ave
r1, c1 = get_ratio(microphones[0], microphones[2])
r2, c2 = get_ratio(microphones[1], microphones[3])
confidence = c1 * c2
logging.debug(f"Localization confidence: {confidence/resonant.LOCALIZATION_CORRELATION_THRESHOLD}")
if (r1 >= 0 and r2 >= 0):
angle1 = -math.acos(r1) * (180/math.pi) + 225
angle2 = math.acos(r2) * (180/math.pi) + 135
elif (r1 <= 0 and r2 <= 0):
angle1 = math.acos(r1) * (180/math.pi) - 135
angle2 = -math.acos(r2) * (180/math.pi) + 135
elif (r1 > 0 and r2 < 0):
angle1 = math.acos(r1) * (180/math.pi) - 135
angle2 = math.acos(r2) * (180/math.pi) + 135
elif (r1 < 0 and r2 > 0):
angle1 = -math.acos(r1) * (180/math.pi) + 225
angle2 = -math.acos(r2) * (180/math.pi) + 135
if (angle1 < 0):
angle1 += 360
if (angle2 < 0):
angle2 += 360
ave_angle = ave_angle(angle1, angle2)
source = Source((ave_angle + resonant.ANGLE_OFFSET) % 360, microphones[0].signal)
logging.debug(f"Calculated source: {source}")
print(f"Calculated source: {source} Localization confidence: {confidence}")
if confidence >= resonant.LOCALIZATION_CORRELATION_THRESHOLD:
return source
else:
print('Source ignored')
return None
def test_cycles_reset(sources: List[Source]):
ml = MLStub()
ml.analyze_name = "dog"
scheduler = SourceScheduler(ml)
scheduler.sources = sources
new_src = Source(33, np.array([1, 1, 1, 1, 1, 1]))
source = sources[0]
source.cycles_lived = 2
scheduler.update_equiv_src(new_src)
assert source.cycles_lived == 0
def test_equiv_src_new(sources):
# This has the sources not machine learning ready
ml = MLStub()
scheduler = SourceScheduler(ml)
scheduler.sources = sources
new_src = Source(38.1, np.array([1, 1, 1, 1]))
scheduler.update_equiv_src(new_src)
assert ml.analyzed is False
assert new_src.id != -1
assert new_src in scheduler.sources
def test_equiv_src_existing(sources):
# This has the sources not machine learning ready
resonant.MIN_ML_SAMPLES = 10
ml = MLStub()
scheduler = SourceScheduler(ml)
scheduler.sources = sources
new_src = Source(33, np.array([1, 1, 1, 1]))
scheduler.update_equiv_src(new_src)
assert ml.analyzed is False
assert new_src.id == None
assert new_src not in scheduler.sources
def analyze(self, src: Source):
self.analyzed = True
src.name = self.analyze_name