def classifi(src_file):
_fix_len = 22050 * 2 * 180
flen = os.path.getsize(os.path.join(TEMP_DIR, "temp.wav"))
_my_dict = {}
for _name in GENRE_LIST:
_my_dict[_name] = 0
if flen > _fix_len:
os.system('sox "%s/temp.wav" "%s/output.wav" trim 0 60 : newfile : restart' % (TEMP_DIR, TEMP_DIR))
for _i in range(1, (flen / _fix_len + 1)):
wav_fn = "%s/output%03d.wav" % (TEMP_DIR, _i)
ceps_fn = "%s/output%03d.ceps.npy" % (TEMP_DIR, _i)
ceps.create_ceps(wav_fn)
_my_dict[train_model(create_model, X, y, ceps_fn)] += 1
_max = -1
_dst = ""
for _key in _my_dict:
if _my_dict[_key] > _max:
_max = _my_dict[_key]
_dst = _key
else:
wav_fn = "%s/temp.wav" % TEMP_DIR
ceps_fn = "%s/temp.ceps.npy" % TEMP_DIR
ceps.create_ceps(wav_fn)
_dst = train_model(create_model, X, y, ceps_fn)
_cmd = 'mv "%s" "%s/%s"' % (src_file, DEST_DIR, _dst)
print(">>>[%s]" % _cmd)
os.system(_cmd)
def do_file(p_path,path):
print(">>> [%s/%s] :" % (p_path,path))
f = os.popen("ls %s/%s"%(p_path,path))
fs = f.read()
f.close()
f_list = fs.split('\n')
print(">>> [%s]" % f_list)
for _f in f_list[:-1]:
if _f.find('.wav')>0:
print('>>> wav file: [%s]' % _f)
fn = '%s/%s/%s' % (p_path,path,_f)
print('--- [%s]' % fn)
ceps.create_ceps(fn)
def test_model_on_single_file(file_path):
clf = joblib.load('data/model_ceps.pkl')
create_ceps(test_file)
fnpy = file_path[:-3]+"ceps.npy"
print fnpy
X, y = read_ceps_test(fnpy)
probs = clf.predict_proba(X)
print str(probs)
#print "\t".join(str(x) for x in traverse)
#print "\t".join(str("%.3f" % x) for x in probs[0])
probs=probs[0]
max_prob = max(probs)
for i,j in enumerate(probs):
if probs[i] == max_prob:
max_prob_index=i
print max_prob_index
predicted_genre = genre_list[max_prob_index]
print "\n\npredicted genre = ",predicted_genre
return predicted_genre
def classifi(src_file):
''' 获取音频文件长度 '''
flen = os.path.getsize(os.path.join(TEMP_DIR,'temp.wav'))
print(">>> flen=%d" % flen)
_my_dict = {}
for _name in GENRE_LIST:
_my_dict[_name] = 0
''' 将大音频文件（大于3分钟的）分割成3分钟采样率22050的单声道音频文件 '''
_fix_len = 22050 * 3 * 60
n = flen/_fix_len
print(">>> n=%d" % n)
if flen>_fix_len:
''' 将大音频文件拆分成小样本 '''
os.system(r'sox "%s"/temp.wav "%s"/output.wav trim 0 180 : newfile : restart' %
(TEMP_DIR,TEMP_DIR))
wav_fn = '%s/output001.wav' % TEMP_DIR
flen = os.path.getsize(wav_fn)
for _i in range(1,n):
'''对拆分音频进行分类
'''
wav_fn = '%s/output%03d.wav' % (TEMP_DIR,_i)
if flen==os.path.getsize(wav_fn):
ceps_fn = '%s/output%03d.ceps.npy' % (TEMP_DIR,_i)
#print(">>> ...[%s]" % ceps_fn)
'''创建测试样本
'''
ceps.create_ceps(wav_fn)
'''样本分类，并分类积分
'''
_my_dict[train_model(create_model, X, y, ceps_fn)]+=1
'''大数判决
'''
_max = -1
_dst= ''
for _key in _my_dict:
if _my_dict[_key]>_max:
_max = _my_dict[_key]
_dst = _key
else:
wav_fn = '%s/temp.wav' % TEMP_DIR
ceps_fn = '%s/temp.ceps.npy' % TEMP_DIR
'''创建测试样本
'''
ceps.create_ceps(wav_fn)
'''样本分类
'''
_dst = train_model(create_model, X, y, ceps_fn)
''' 将原始的音频文件 放置 到归类的目录下
2014.12.14：问题，若在目录下有同名文件，结果？
－增加时间戳
'''
_t = str(time.time())
_cmd = 'mv "%s" "%s"/%s/"%s"' % (src_file,DEST_DIR,_dst,(_t+'_'+os.path.split(src_file)[1]))
print(_cmd)
os.system(_cmd)
import sys
from save import make_wav
from ceps import create_ceps
if __name__ == "__main__":
DIR = "/home/viewtiful/2018-cap1-7/src/mfcc/"
# DIR = "C:\Users\lynn\PycharmProjects\\2018-cap1-7\src\mfcc"
f = sys.argv[1]
while 1:
num = int(input())
sec = int(input())
make_wav(f, num, sec)
fn = DIR + f + str(num) + ".wav"
X = []
create_ceps(fn)