def create_annotation(root, annotator_id, jam_file, context):
"""Creates an annotation from the given root of an XML svl file."""
# Load jam file
jam = jams2.load(jam_file)
# If annotation exists, replace it
annot = None
for section in jam.sections:
if section.annotation_metadata.annotator.name == annotator_id:
annot = section
# If this context already exists, do nothing
for data in annot.data:
if data.label.context == context:
return
break
# Create Annotation if needed
if annot is None:
annot = jam.sections.create_annotation()
# Create Metadata
annot.annotation_metadata.annotator = annotators[annotator_id]
# TODO: More metadata
# Get sampling rate from XML root
sr = float(root.iter("model").next().attrib["sampleRate"])
# Create datapoints from the XML root
points = root.iter("point")
point = points.next()
start = float(point.attrib["frame"]) / sr
label = point.attrib["label"]
for point in points:
section = annot.create_datapoint()
section.start.value = start
section.end.value = float(point.attrib["frame"]) / sr
# Make sure upper and lower case are consistent
if context == "small_scale":
section.label.value = label.lower()
elif context == "large_scale":
section.label.value = label.upper()
section.label.context = context
start = float(point.attrib["frame"]) / sr
label = point.attrib["label"]
# Save file
with open(jam_file, "w") as f:
json.dump(jam, f, indent=2)
def process_track(in_path, audio_file, jam_file, annot_beats, feature, ds_name,
annot_bounds, rank, h, R):
# Only analize files with annotated beats
if annot_beats:
jam = jams2.load(jam_file)
if jam.beats == []:
return
if jam.beats[0].data == []:
return
logging.info("Segmenting %s" % audio_file)
# C-NMF segmenter call
if rank is None:
est_times, est_labels = S.process(audio_file,
feature=feature,
annot_beats=annot_beats,
annot_bounds=annot_bounds)
else:
est_times, est_labels = S.process(audio_file,
feature=feature,
annot_beats=annot_beats,
annot_bounds=annot_bounds,
rank=rank,
h=h,
R=R)
# Save
out_file = os.path.join(in_path, "estimations",
os.path.basename(audio_file)[:-4] + ".json")
if not annot_bounds:
MSAF.save_estimations(out_file,
est_times,
annot_beats,
"cnmf2",
bounds=True,
feature=feature)
MSAF.save_estimations(out_file,
est_labels,
annot_beats,
"cnmf2",
bounds=False,
annot_bounds=annot_bounds,
feature=feature)
return []
def compute_gt_results(est_file,
trim,
annot_beats,
jam_files,
alg_id,
beatles=False,
annotator=0,
bins=10,
**params):
"""Computes the results by using the ground truth dataset."""
# Get the ds_prefix
ds_prefix = os.path.basename(est_file).split("_")[0]
# Get corresponding annotation file
jam_file = get_annotation(est_file, jam_files)
if beatles:
jam = jams2.load(jam_file)
if jam.metadata.artist != "The Beatles":
return []
try:
if annotator == "GT" or annotator == 0:
ann_inter, ann_labels = jams2.converters.load_jams_range(
jam_file,
"sections",
annotator=0,
context=MSAF.prefix_dict[ds_prefix])
else:
ann_inter, ann_labels = jams2.converters.load_jams_range(
jam_file,
"sections",
annotator_name=annotator,
context="large_scale")
except:
logging.warning("No annotations for file: %s" % jam_file)
return []
est_inter = MSAF.read_estimations(est_file, alg_id, annot_beats, **params)
if est_inter == []:
return []
# Compute the results and return
return compute_results(ann_inter, est_inter, trim, bins, est_file)
def enrich_jam(audio_file, jam_file):
"""Enriches the JAMS file with the audio file."""
# Read JAMS and annotation
jam = jams2.load(jam_file)
# Read Echo Nest Info
#while True:
#try:
#pytrack = track.track_from_filename(audio_file)
#break
#except:
#logging.warning("Connection lost. Retrying in 5 seconds...")
#time.sleep(5)
## Fill data
#dur = pytrack.duration
#jam.metadata.duration = float(pytrack.duration)
#jam.metadata.md5 = pytrack.md5
#jam.metadata.echonest_id = pytrack.id
#try:
#jam.metadata.artist = pytrack.artist
#except AttributeError:
#pass
# Use sox instead of Echo Nest
parsed_audio_file = audio_file.replace(",", "\,")
parsed_audio_file = parsed_audio_file.replace(" ", "\ ")
parsed_audio_file = parsed_audio_file.replace("&", "\&")
parsed_audio_file = parsed_audio_file.replace("'", "\\'")
parsed_audio_file = parsed_audio_file.replace(")", "\)")
parsed_audio_file = parsed_audio_file.replace("(", "\(")
cmd = "soxi -D " + parsed_audio_file
dur = float(subprocess.check_output(cmd, shell=True))
jam.metadata.duration = dur
# Add the last segment
if os.path.basename(audio_file)[:9] == "Cerulean_":
add_last_segment(jam, jam_file, dur)
# Save JAMS
with open(jam_file, "w") as f:
json.dump(jam, f, indent=2)