class SoundRecognitionApp():
def __init__(self, cfg) -> None:
self.transformer = Signal2ImageTransformer(**cfg['transforms'])
self.audio = Audio(cfg['audio'])
self.load_model(cfg['model'])
pass
def run(self):
print("============= REALTIME START ==============")
self.audio.start()
self.flag = True
try:
while self.flag:
status, data = self.audio.get()
if status == Audio.ERROR:
print('[error]')
break
elif status == Audio.WAIT:
continue
mel_spec = self.preprocess(data)
result = self.inference(mel_spec)
except KeyboardInterrupt:
pass
except Exception as e:
print(e)
finally:
self.audio.stop()
print("============= REALTIME FINISH ==============")
def preprocess(self, signal):
return np.expand_dims(self.transformer.transform(signal), axis=0)
def inference(self, X):
image = torch.from_numpy(X.astype(np.float32)).clone()
image.to(self.device).float()
prob = self.model(image)['multilabel_proba'].detach().cpu().numpy()
return prob
def load_model(self, cfg):
try:
self.device = torch.device(cfg["device"])
self.model = getattr(ml.my_model, cfg['name'])(**cfg['params'])
self.model.load_state_dict(torch.load(cfg['path']))
self.model.to(self.device)
except AttributeError as e:
print(f"Model {cfg['name']} is None. {e}")
exit(1)
except FileNotFoundError as e:
print(f"{e}")
exit(1)
except Exception as e:
print(f"{e}")
exit(1)
class MusicChangePointDetector(object):
def __init__(self, setting_path: str, audio_path: str):
"""コンストラクタ
Args:
setting_path (str): 設定ファイルのパス
audio_path (str): 音楽ファイルのパス
"""
with open(setting_path, 'r') as f:
cfg = yaml.load(f)
self.cf = cf.ChangeFinder(**cfg['change_finder'])
self.audio = Audio(cfg['audio'], audio_file_path=audio_path)
self.buffer = np.zeros(cfg['model']['buffer_audio_length'],
dtype=np.float32)
self.buf_num = int(cfg['model']['frame_buf_num'])
self.spec_buf = []
self.thr = float(cfg['model']['thr'])
def run(self):
"""メインループ開始"""
self.audio.start()
try:
while True:
status, data = self.audio.get()
if status == Audio.ERROR:
break
elif status == Audio.WAIT:
continue
self.buffer = np.roll(self.buffer, -data.shape[0], axis=0)
self.buffer[-data.shape[0]:] = data
if self.detect():
print('detect')
except KeyboardInterrupt:
print('Interrupt')
self.audio.stop()
def detect(self):
"""検出
Returns:
bool: 転調したかどうか
"""
is_detect = False
D = np.average(librosa.amplitude_to_db(np.abs(librosa.stft(
self.buffer)),
ref=np.max),
axis=1)[:512]
D /= np.linalg.norm(D, ord=2)
self.spec_buf.append(D)
if len(self.spec_buf) > self.buf_num:
similarity = np.average(
np.dot(self.spec_buf[-1],
np.array(self.spec_buf[-(self.buf_num - 1):-1]).T))
score = self.cf.update(similarity)
self.spec_buf.pop()
if score > self.thr:
is_detect = True
return is_detect
