def test_labels():
# For multiple tags
default_clipid = "64760"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
tags = clip.tags
assert tags.labels == [
"Electric_guitar",
"Guitar",
"Plucked_string_instrument",
"Musical_instrument",
"Music",
]
assert np.array_equal(tags.confidence, [1.0, 1.0, 1.0, 1.0, 1.0])
mids = clip.mids
assert mids.labels == [
"/m/02sgy", "/m/0342h", "/m/0fx80y", "/m/04szw", "/m/04rlf"
]
assert np.array_equal(mids.confidence, [1.0, 1.0, 1.0, 1.0, 1.0])
# For a single tag
default_clipid = "21914"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
tags = clip.tags
assert tags.labels == ["Crushing"]
assert np.array_equal(tags.confidence, [1.0])
def test_vocabularies():
dataset = fsd50k.Dataset(TEST_DATA_HOME)
# Testing load vocabulary function
fsd50k_to_audioset, audioset_to_fsd50k = dataset.load_fsd50k_vocabulary(
dataset.vocabulary_path
)
assert fsd50k_to_audioset == {
"Crushing": "/m/07plct2",
"Electric_guitar": "/m/02sgy",
}
assert audioset_to_fsd50k == {
"/m/07plct2": "Crushing",
"/m/02sgy": "Electric_guitar",
}
# Testing fsd50k to audioset
fsd50k_to_audioset = dataset.fsd50k_to_audioset
assert fsd50k_to_audioset["Crushing"] == "/m/07plct2"
assert fsd50k_to_audioset["Electric_guitar"] == "/m/02sgy"
# Testing audioset to fsd50k
audioset_to_fsd50k = dataset.audioset_to_fsd50k
assert audioset_to_fsd50k["/m/07plct2"] == "Crushing"
assert audioset_to_fsd50k["/m/02sgy"] == "Electric_guitar"
def test_load_audio():
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip("64760")
audio_path = clip.audio_path
audio, sr = fsd50k.load_audio(audio_path)
assert sr == 44100
assert type(audio) is np.ndarray
assert len(audio.shape) == 1 # check audio is loaded as mono
assert len(audio) == 75601
def test_to_jams():
default_clipid = "64760"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
jam = clip.to_jams()
# Validate fsd50k jam schema
assert jam.validate()
# Validate Tags
tags = jam.search(namespace="tag_open")[0]["data"]
assert len(tags) == 5
assert [tag.time for tag in tags] == [0.0, 0.0, 0.0, 0.0, 0.0]
assert [tag.duration for tag in tags] == [
1.7143083900226757,
1.7143083900226757,
1.7143083900226757,
1.7143083900226757,
1.7143083900226757,
]
assert [tag.value for tag in tags] == [
"Electric_guitar",
"Guitar",
"Plucked_string_instrument",
"Musical_instrument",
"Music",
]
assert [tag.confidence for tag in tags] == [1.0, 1.0, 1.0, 1.0, 1.0]
# validate metadata
assert jam.file_metadata.duration == 1.7143083900226757
assert jam.file_metadata.title == "guitarras_63.wav"
assert jam.sandbox.mids == [
"/m/02sgy",
"/m/0342h",
"/m/0fx80y",
"/m/04szw",
"/m/04rlf",
]
assert jam.sandbox.split == "train"
assert jam.sandbox.description == "electric guitar"
assert jam.sandbox.freesound_tags == [
"electric",
"guitar",
]
assert jam.sandbox.license == "http://creativecommons.org/licenses/sampling+/1.0/"
assert jam.sandbox.uploader == "casualsamples"
assert jam.sandbox.pp_pnp_ratings == {"/m/02sgy": [1.0, 1.0]}
assert jam.annotations[0].annotation_metadata.data_source == "soundata"
def test_label_info():
dataset = fsd50k.Dataset(TEST_DATA_HOME)
# Testing label info property
label_info = dataset.label_info
assert type(label_info["/m/02sgy"]) is dict
assert type(label_info["/m/02sgy"]["faq"]) is str
assert label_info["/m/02sgy"]["examples"] == [4282, 134012]
assert label_info["/m/02sgy"]["verification_examples"] == [
63900,
74871,
40403,
97244,
40474,
97242,
]
def test_collection_vocabulary():
dataset = fsd50k.Dataset(TEST_DATA_HOME)
# Testing collection vocabularies
collection_fsd50k_to_audioset = dataset.collection_fsd50k_to_audioset
collection_audioset_to_fsd50k = dataset.collection_audioset_to_fsd50k
assert type(collection_fsd50k_to_audioset) is dict
assert type(collection_audioset_to_fsd50k) is dict
assert type(collection_fsd50k_to_audioset["dev"]) is dict
assert type(collection_fsd50k_to_audioset["eval"]) is dict
assert type(collection_audioset_to_fsd50k["dev"]) is dict
assert type(collection_audioset_to_fsd50k["eval"]) is dict
assert collection_fsd50k_to_audioset["dev"]["Electric_guitar"] == "/m/02sgy"
assert collection_fsd50k_to_audioset["eval"]["Chatter"] == "/m/07rkbfh"
assert collection_audioset_to_fsd50k["dev"]["/m/02sgy"] == "Electric_guitar"
assert collection_audioset_to_fsd50k["eval"]["/m/07rkbfh"] == "Chatter"
def test_clip():
default_clipid = "64760"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
expected_attributes = {
"audio_path": "tests/resources/sound_datasets/fsd50k/FSD50K.dev_audio/64760.wav",
"clip_id": "64760",
}
expected_property_types = {
"audio": tuple,
"tags": annotations.Tags,
"mids": annotations.Tags,
"split": str,
"title": str,
"description": str,
"pp_pnp_ratings": dict,
}
run_clip_tests(clip, expected_attributes, expected_property_types)
def test_dev_metadata():
# Testing metadata from a training clip
default_clipid = "64760"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
clip_metadata = clip._metadata()
clip_ground_truth = clip_metadata[default_clipid]["ground_truth"]
assert clip_ground_truth["tags"] == [
"Electric_guitar",
"Guitar",
"Plucked_string_instrument",
"Musical_instrument",
"Music",
]
assert clip_ground_truth["mids"] == [
"/m/02sgy",
"/m/0342h",
"/m/0fx80y",
"/m/04szw",
"/m/04rlf",
]
assert clip_ground_truth["split"] == "train"
clip_info = clip_metadata[default_clipid]["clip_info"]
assert clip_info["title"] == "guitarras_63.wav"
assert clip_info["description"] == "electric guitar"
assert clip_info["tags"] == ["electric", "guitar"]
assert clip_info[
"license"] == "http://creativecommons.org/licenses/sampling+/1.0/"
assert clip_info["uploader"] == "casualsamples"
clip_pp_pnp = clip_metadata[default_clipid]["pp_pnp_ratings"]
assert type(clip_pp_pnp) is dict
assert clip_pp_pnp == {"/m/02sgy": [1.0, 1.0]}
# Testing metadata from an evaluation clip
default_clipid = "21914"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
clip_metadata = clip._metadata()
clip_ground_truth = clip_metadata[default_clipid]["ground_truth"]
assert clip_ground_truth["split"] == "validation"
default_clipid = "99"
dataset = fsd50k.Dataset(TEST_DATA_HOME)
clip = dataset.clip(default_clipid)
clip_metadata = clip._metadata()
clip_ground_truth = clip_metadata[default_clipid]["ground_truth"]
assert clip_ground_truth["tags"] == [
"Chatter",
"Chewing_and_mastication",
"Chirp_and_tweet",
"Traffic_noise_and_roadway_noise",
"Child_speech_and_kid_speaking",
"Human_group_actions",
"Bird_vocalization_and_bird_call_and_bird_song",
"Bird",
"Wild_animals",
"Animal",
"Motor_vehicle_(road)",
"Vehicle",
"Speech",
"Human_voice",
]
assert clip_ground_truth["mids"] == [
"/m/07rkbfh",
"/m/03cczk",
"/m/07pggtn",
"/m/0btp2",
"/m/0ytgt",
"/t/dd00012",
"/m/020bb7",
"/m/015p6",
"/m/01280g",
"/m/0jbk",
"/m/012f08",
"/m/07yv9",
"/m/09x0r",
"/m/09l8g",
]
assert clip_ground_truth["split"] is "test"
clip_info = clip_metadata[default_clipid]["clip_info"]
assert type(clip_info) is dict
assert clip_info["title"] == "manzana_exterior.wav"
assert clip_info["description"] == "eating apples on a park"
assert clip_info["tags"] == [
"apple", "crunch", "eat", "field-recording", "park"
]
assert clip_info[
"license"] == "http://creativecommons.org/licenses/by/3.0/"
assert clip_info["uploader"] == "plagasul"
clip_pp_pnp = clip_metadata[default_clipid]["pp_pnp_ratings"]
assert type(clip_pp_pnp) is dict
assert clip_pp_pnp == {"/m/03cczk": [0.5, 0.5]}