def test_source_wet_is_delayed_by_one_k():
env = CrossAdaptiveEnv()
action = env.action_space.sample()
N = 2
for _ in range(N):
env.step(action)
assert env.source_dry.player.k == N
assert env.target.player.k == N
assert env.source_wet.player.k == N - 1
def test_non_debug_mode_does_not_define_debug_channels():
config = {
**CROSS_ADAPTIVE_DEFAULT_CONFIG,
"debug": False,
}
env = CrossAdaptiveEnv(config)
debug_channels = list(
map(lambda param: f"{param.name}{DEBUG_SUFFIX}",
env.effect.parameters))
action = env.action_space.sample()
env.step(action)
source = env.render()
for ch in debug_channels:
assert f'chn_k "{ch}"' not in source.csd
def test_debug_mode_renders_channels_to_debug_wave_file():
config = {
**CROSS_ADAPTIVE_DEFAULT_CONFIG,
"debug": True,
}
env = CrossAdaptiveEnv(config)
debug_channels = list(
map(lambda param: f"{param.name}{DEBUG_SUFFIX}",
env.effect.parameters))
action = env.action_space.sample()
env.step(action)
source = env.render()
assert "fout" in source.csd
for ch in debug_channels:
assert f"upsamp(k_{ch})" in source.csd
def test_debug_mode_sets_debug_channels():
config = {
**CROSS_ADAPTIVE_DEFAULT_CONFIG,
"debug": True,
}
env = CrossAdaptiveEnv(config)
debug_channels = list(
map(lambda param: f"{param.name}{DEBUG_SUFFIX}",
env.effect.parameters))
action = env.action_space.sample()
env.step(action)
source = env.render()
debug_values = source.player.get_channels(debug_channels)
for v in debug_values:
assert env.action_space.low[0] < v < env.action_space.high[1]
def test_source_wet_wraps_correctly_at_the_end_of_the_sound():
config = {
**CROSS_ADAPTIVE_DEFAULT_CONFIG,
"eval_interval": None, # episode is done at the end of the source
}
env = CrossAdaptiveEnv(config)
action = env.action_space.sample()
assert env.should_delay_source_wet_one_frame is True
done = False
while not done:
_, _, done, _ = env.step(action)
assert env.source_dry.player.k == 0
assert env.source_wet.player.k > 0
assert env.should_delay_source_wet_one_frame is False
_, _, done, _ = env.step(action)
assert env.source_wet.player.k == 0
assert env.source_dry.player.k == 1
def test_source_wet_is_equal_to_previous_source_dry_when_effect_is_thru():
config = {**CROSS_ADAPTIVE_DEFAULT_CONFIG, "effect": "thru"}
env = CrossAdaptiveEnv(config)
env.step(env.action_space.sample())
source_dry_features_after_first_step = env.source_dry_features.copy()
source_wet_features_after_first_step = env.source_wet_features.copy()
assert np.array_equal(
source_wet_features_after_first_step,
np.zeros(shape=len(env.analyser.analysis_features)),
)
env.step(env.action_space.sample())
source_dry_features_after_second_step = env.source_dry_features.copy()
source_wet_features_after_second_step = env.source_wet_features.copy()
assert np.array_equal(source_wet_features_after_second_step,
source_dry_features_after_first_step)
assert not np.array_equal(source_dry_features_after_second_step,
source_dry_features_after_first_step)
def run_offline_inference(agent: Trainer, env: CrossAdaptiveEnv):
# NOTE: something is wrong here. For some reason, all the action values are too close to the bound
done = False
obs = env.reset()
while not done:
action = agent.compute_action(obs)
# TODO: standardize action
# it might be difficult to standardize the action in live mode, but offline inference essentially work
obs, _, done, _ = env.step(action)
def test_multiple_targets_are_allowed():
targets = ["amen_5s.wav", "drums_5s.wav"]
config = {
**CROSS_ADAPTIVE_DEFAULT_CONFIG.copy(),
"targets": targets,
"eval_interval": None,
}
env = CrossAdaptiveEnv(config)
assert Path(env.target.input).name == targets[0]
done = False
while not done:
action = env.action_space.sample()
_, _, done, _ = env.step(action)
assert Path(env.target.input).name == targets[1]
done = False
while not done:
action = env.action_space.sample()
_, _, done, _ = env.step(action)
assert Path(env.target.input).name == targets[0]
def test_env_inits_and_makes_first_step_correctly():
env = CrossAdaptiveEnv()
empty_features = np.zeros(shape=len(env.analyser.analysis_features))
initial_state = env.get_state()
assert np.array_equal(initial_state,
np.concatenate((empty_features, empty_features)))
action = env.action_space.sample()
state, reward, done, _ = env.step(action)
assert done is True
assert reward == 0.0
assert not np.array_equal(state, initial_state)
assert np.abs(state.mean()) > 0.0