def test_oscillator_called_with_correct_params_when_freq_is_scalar(
self, mocker
):
dummy_operator = Sinusoid()
mocker.patch.object(
dummy_operator, "forward", return_value=torch.Tensor([4, 1, 2, 3])
)
operators = (dummy_operator,)
envelope_generators = (ADSR(),)
operator_routing = ()
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([1, 1, 1, 1])
frequency = torch.Tensor([1234])
ratios = (1,)
a, d, s, r = 0.9, 0.8, 0.7, 0.6
fm_synth(gate, frequency, ratios, ((a, d, s, r),))
args = dummy_operator.forward.call_args[0]
kwargs = dummy_operator.forward.call_args[1]
torch.testing.assert_allclose(
args[0], frequency.repeat_interleave(gate.shape[-1], dim=-1)
)
assert kwargs["sample_rate"] == sample_rate
assert kwargs["phase_mod"] is None
def test_oscillator_frequency_adjusted_by_ratio(self, mocker):
operators = (Sinusoid(), Sinusoid())
op0_dummy_return = torch.Tensor([1, 2, 3, 4])
op1_dummy_return = torch.Tensor([4, 3, 2, 1])
mocker.patch.object(
operators[0], "forward", return_value=op0_dummy_return
)
mocker.patch.object(
operators[1], "forward", return_value=op1_dummy_return
)
envelope_generators = (ADSR(), ADSR())
operator_routing = ()
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([1, 1, 1, 1])
frequency = torch.Tensor([1, 1, 1, 1])
ratios = (1, 4)
a, d, s, r = 0.9, 0.8, 0.7, 0.6
fm_synth(gate, frequency, ratios, ((a, d, s, r), (a, d, s, r)))
torch.testing.assert_allclose(
operators[0].forward.call_args[0][0], frequency * ratios[0],
)
torch.testing.assert_allclose(
operators[1].forward.call_args[0][0], frequency * ratios[1],
)
def test_envelope_generator_called_with_correct_params(self, mocker):
operators = (Sinusoid(),)
dummy_adsr = ADSR()
mocker.patch.object(
dummy_adsr, "forward", return_value=torch.Tensor([1, 2, 3, 4])
)
envelope_generators = (dummy_adsr,)
operator_routing = ()
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([0, 1, 0, 1])
frequency = torch.Tensor([1000])
ratios = (1,)
a, d, s, r = 0.1, 0.2, 0.3, 0.4
fm_synth(gate, frequency, ratios, ((a, d, s, r),))
dummy_adsr.forward.assert_called_once_with(
gate, a, d, s, r, sample_rate
)
def test_oscillator_called_with_correct_params_when_freq_is_tensor(
self, mocker
):
dummy_operator = Sinusoid()
mocker.patch.object(
dummy_operator,
"forward",
return_value=torch.Tensor([1, 2, 3, 4, 5]),
)
operators = (dummy_operator,)
envelope_generators = (ADSR(),)
operator_routing = ()
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([1, 0, 1, 1, 0])
frequency = torch.Tensor([1234, 1234, 1234, 4321, 0])
ratios = (1,)
a, d, s, r = 0.9, 0.8, 0.7, 0.6
fm_synth(gate, frequency, ratios, ((a, d, s, r),))
torch.testing.assert_allclose(
dummy_operator.forward.call_args[0][0], frequency
)
def test_can_construct_with_correct_interface(self):
operators = (Sinusoid(), Sinusoid(), Sinusoid(), Sinusoid())
envelope_generators = (ADSR(), ADSR(), ADSR(), ADSR())
operator_routing = ((3, 0), (2, 1), (1, 0))
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
del fm_synth
def test_operators_correctly_chained_in_complex_config(self, mocker):
op_dummy_returns = [torch.Tensor([n, n, n]) for n in range(5)]
operators = [Sinusoid() for _ in range(5)]
for dummy_return, operator in zip(op_dummy_returns, operators):
mocker.patch.object(operator, "forward", return_value=dummy_return)
eg_dummy_returns = [torch.Tensor([0, 1, 0]) for _ in range(5)]
envelope_generators = [ADSR() for _ in range(5)]
for dummy_return, eg in zip(eg_dummy_returns, envelope_generators):
mocker.patch.object(eg, "forward", return_value=dummy_return)
operator_routing = ((1, 0), (2, 0), (3, 2), (3, 1), (1, 4))
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([1, 1, 1])
frequency = torch.Tensor([1])
ratios = (1, 2, 3.5, 0.1, 2.2)
a, d, s, r = 0.9, 0.8, 0.7, 0.6
expected_output = (
eg_dummy_returns[0] * op_dummy_returns[0]
+ eg_dummy_returns[4] * op_dummy_returns[4]
)
actual_output = fm_synth(gate, frequency, ratios, [(a, d, s, r)] * 5)
op_kwargs = [operator.forward.call_args[1] for operator in operators]
torch.testing.assert_allclose(
op_kwargs[0]["phase_mod"],
(
eg_dummy_returns[1] * op_dummy_returns[1]
+ eg_dummy_returns[2] * op_dummy_returns[2]
),
)
torch.testing.assert_allclose(
op_kwargs[1]["phase_mod"],
(eg_dummy_returns[3] * op_dummy_returns[3]),
)
torch.testing.assert_allclose(
op_kwargs[2]["phase_mod"],
(eg_dummy_returns[3] * op_dummy_returns[3]),
)
torch.testing.assert_allclose(
op_kwargs[4]["phase_mod"],
(eg_dummy_returns[1] * op_dummy_returns[1]),
)
torch.testing.assert_allclose(actual_output, expected_output)
def test_operators_correctly_chained_in_simple_config(self, mocker):
op0_dummy_return = torch.Tensor([1, 2, 3, 4])
op1_dummy_return = torch.Tensor([4, 3, 2, 1])
operators = (Sinusoid(), Sinusoid())
mocker.patch.object(
operators[0], "forward", return_value=op0_dummy_return
)
mocker.patch.object(
operators[1], "forward", return_value=op1_dummy_return
)
eg0_dummy_return = torch.Tensor([10, 10, 10, 10])
eg1_dummy_return = torch.Tensor([-1, -1, -1, -1])
envelope_generators = (ADSR(), ADSR())
mocker.patch.object(
envelope_generators[0], "forward", return_value=eg0_dummy_return,
)
mocker.patch.object(
envelope_generators[1], "forward", return_value=eg1_dummy_return,
)
operator_routing = ((1, 0),)
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([1, 1, 1, 1])
frequency = torch.Tensor([1, 1, 1, 1])
ratios = (1, 4)
op_gains = (torch.Tensor([0.5, 0.5, 0.5, 0.5]), 0.8)
a, d, s, r = 0.9, 0.8, 0.7, 0.6
expected_output = eg0_dummy_return * op0_dummy_return * op_gains[0]
actual_output = fm_synth(
gate, frequency, ratios, ((a, d, s, r), (a, d, s, r)), op_gains
)
op0_kwargs = operators[0].forward.call_args[1]
torch.testing.assert_allclose(
op0_kwargs["phase_mod"],
op1_dummy_return * eg1_dummy_return * op_gains[1],
)
torch.testing.assert_allclose(actual_output, expected_output)
def test_can_forward_with_correct_interface(self):
operators = (Sinusoid(), Sinusoid())
operator_routing = ((1, 0),)
envelope_generators = (ADSR(), ADSR())
sample_rate = 44100
fm_synth = FMSynth(
operators, envelope_generators, operator_routing, sample_rate
)
gate = torch.Tensor([0, 0, 0, 1, 1, 1, 0, 0, 0])
frequency = torch.Tensor([1])
ratios = (1, 2)
a_0, d_0, s_0, r_0 = 0.3, 0.2, 0.5, 0.1
a_1, d_1, s_1, r_1 = 0.3, 0.2, 0.5, 0.1
fm_synth(
gate,
frequency,
ratios,
((a_0, d_0, s_0, r_0), (a_1, d_1, s_1, r_1)),
)