def read_from_mp3_folder(path):
"""
Reads folder of mp3s into database
Parameters
----------
path : String
global path to folder
Returns
-------
Database
database of songs
Dependencies
------------
Song to fingerprint
append_database
"""
database = Database()
folder = Path(path)
for file in folder.iterdir():
if not file.is_dir() and file.suffix == ".mp3":
print(file.stem)
name = file.stem
samples = get_mp3_data(file)
peaks = sample_to_peaks(samples)
fingerprint = get_fingerprint(peaks, 15, len(database.id_to_name))
append_database(database, fingerprint, name)
return database
def get_post_upload():
"""
Function is called when a user wants to upload sound and meta-data
to the database.
"""
file_key = "sound_file"
upload_folder = "sounds"
if request.method == "POST" and file_key in request.files:
sound_file = request.files[file_key]
file_path = os.path.join(
upload_folder,
request.form["sound_name"] + ".wav"
)
sound_file.save(file_path)
db.insert_reference(
request.form["sound_name"],
fingerprint.get_fingerprint(file_path),
float(request.form["r_ref"]),
float(request.form["l_ref"]),
request.form["class"]
)
return render_template("upload.html")
def get_post_upload():
"""
Function is called when a user wants to upload sound and meta-data
to the database.
"""
file_key = "sound_file"
upload_folder = "sounds"
if request.method == "POST" and file_key in request.files:
sound_file = request.files[file_key]
file_path = os.path.join(upload_folder,
request.form["sound_name"] + ".wav")
sound_file.save(file_path)
db.insert_reference(request.form["sound_name"],
fingerprint.get_fingerprint(file_path),
float(request.form["r_ref"]),
float(request.form["l_ref"]),
request.form["class"])
return render_template("upload.html")
def init():
"""
Initializes the RethinkDB database
@return Whether the database was reinitialized or not
"""
conn = r.connect()
if not DB in r.db_list().run(conn):
create(conn)
f_print = fingerprint.get_fingerprint(SAMPLE_PATH)
insert_reference(
SAMPLE_NAME,
f_print,
SAMPLE_R_REF,
SAMPLE_L_REF,
SAMPLE_CLASS
)
return True
else:
return False
def get_fingerprint(file_name):
f_new = converter.convert(file_name)
fp = fingerprint.get_fingerprint(f_new)
os.remove(os.path.join(cf.FOLDER_TEMP, f_new))
return fp
def check_matches(audio_data, database):
"""
Checks if audio data from a song matches a song in the database, and if so, which song it is
Parameters
----------
audio_data : np.array([])
A 1D array of audio samples from recorded audio or audio file.
database : dictionary
An dictionary containing fingerprint (peak to list of songs) mappings for all songs
Returns
-------
int()
The song id of the recognized song; -1 if no song recognized
Dependencies
------------
sample_to_spectrogram()
spectrogram_to_peaks()
<peaks to keys> (to be named)
"""
req = 20 #Determine through experimentation
no_match = -1
kvpairs = fp.get_fingerprint(spk.sample_to_peaks(audio_data), 15)
#audio_data, times --> spectrogram --> array of peaks: peak = (t, f) --> list of peaks in the song
#COUNT NUMBER OF MATCHED PEAKS FOR EACH SONG IN THE DATABASE, RECORD OFFSETS
match_cnt = Counter()
#print(database)
#print(kvpairs)
for k, v in kvpairs.items():
#print(k)
if k in database:
#print(k)
#print(database[k])
for value in database[k]:
#print("value: {}".format(value))
#print(v)
match_cnt.update(((value[0], value[1] - v[0][1]), ))
#ELIMINATE SONGS WITH INSUFFICIENT MATCHES
bad_matches = []
for match in match_cnt:
#print(match)
if match_cnt[match] < req:
#print(match_cnt[match])
bad_matches.append(match)
for bad_match in bad_matches:
del match_cnt[bad_match]
#print(match_cnt)
if len(match_cnt) == 0:
return no_match
#RETURN THE SONG ID WITH THE MOST MATCHES
return match_cnt.most_common(1)[0][0][0]
#OLD CODE--WILL PROBABLY NOT NEED (IGNORE)
'''
sample_time = np.array(v)
song_time = np.array(database[k])
sample_time[:,1] = song_time[:,1] - sample_time[:,1] #maybe will have to reverse, but this works with array broadcasting
match_cnt.update(sample_time)
'''
'''
for pair in v:
if pair[0] not in match_cnt:
match_cnt[pair[0]] = 1
else:
match_cnt[pair[0]] += 1
'''
'''
times = list() #Times at which peaks were found in given song
posmatches = list() #Values (lists of (song1, time), (song2, time)...) for possible matches
'''
'''
for p, t in kvpairs:
if p in database:
times.append(t) #time at which matched peak was found (in given song) is appended to times
posmatches.append(database[p]) #actual corresponding time and song values from database are recorded
'''
'''
img_collector = collector.ImageCollector()
for filename in imgs:
camera = extract_camera_name(filename)
img_collector[camera].append(filename)
cameras = list(img_collector.imgs.keys())
print('Images loaded!')
# Get fingerprint
print('Generating fingerprints...')
fps = []
labels = []
count = 0
for camera in cameras:
imgs = img_collector.imgs[camera]
for group in chunks(imgs, properties['group']):
fp = fingerprint.get_fingerprint(imgs, camera)
fp = eliminate_nan_inf(fp)
fps.append(fp)
labels.append(camera)
count += 1
print(f"==> {count/len(cameras)}%")
print('Finished!')
# Train NN
print("Training NN...")
shuffled_inds = list(range(len(fps)))
print(np.array(fps).shape)
fps = torch.Tensor(np.transpose(np.array(fps), (0, 3, 1, 2))).to(
device) # Convert to arrays
labels = np.array(labels) # Convert to arrays
random.shuffle(shuffled_inds)
def get_fingerprint(file_name):
f_new = converter.convert(file_name)
fp = fingerprint.get_fingerprint(f_new)
os.remove(os.path.join(cf.FOLDER_TEMP, f_new))
return fp