def process_file(wav_file, class_labels, to_csv, num_files, output_file,
ten_seconds, num_predictions, threshold):
print(wav_file)
files = tf.io.gfile.glob(wav_file)
total_predictions = []
examples = []
print('Number of predictions ' + str(num_predictions))
print('Threshold ' + str(threshold))
if not files:
raise IOError("Unable to find input files. data_pattern='" + wav_file +
"'")
if num_files == 0:
num_files = len(files)
for i in range(0, num_files):
if not files[i]:
i = i + 1
sr, data = wavfile.read(files[i])
if data.dtype != np.int16:
raise TypeError('Bad sample type: %r' % data.dtype)
print(i)
print(files[i])
# local import to reduce start-up time
from audio.processor import WavProcessor, format_predictions
with WavProcessor() as proc:
if ten_seconds == False:
predictions = proc.get_predictions(wav_file, sr, data,
num_predictions, threshold,
class_labels)
print('Predictions')
print(format_predictions(predictions))
if to_csv == True:
examples.append(int(len(data) / 44100))
total_predictions.append(predictions)
else:
predictions = proc.get_predictions2(sr, data, num_predictions,
threshold, class_labels)
print('Predictions')
for i in range(0, len(predictions)):
print(str(i) + ' ' + format_predictions(predictions[i]))
if to_csv == True:
examples.append(int(len(data) / 44100))
total_predictions.append(predictions)
if to_csv == True:
from audio.processor import WavProcessor, format_predictions
with WavProcessor() as proc:
if ten_seconds == False:
proc.toCSV(examples, wav_file, num_files, output_file,
total_predictions)
else:
proc.toCSV2(examples, wav_file, num_files, output_file,
total_predictions)
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data_event.set()
while self.is_running:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data_event.clear()
if self._save_path:
f_path = os.path.join(
self._save_path, 'record_{:.0f}.wav'.format(time.time())
)
wavfile.write(f_path, self._sample_rate, self._process_buf)
logger.info('"{}" saved'.format(f_path))
logger.info('Start processing')
predictions = proc.get_predictions(
self._sample_rate, self._process_buf)
formatted = format_predictions(predictions)
logger.info('Predictions: {}'.format(formatted))
self.events_queue.append((datetime.datetime.now(), formatted))
self._send_dh(predictions)
logger.info('Stop processing')
self._process_buf = None
self._ask_data_event.set()
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data.set()
while True:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data.clear()
if self._save_path:
f_path = os.path.join(
self._save_path,
'record_{:.0f}.wav'.format(time.time()))
wavfile.write(f_path, self._sample_rate, self._process_buf)
logger.info('"{}" saved.'.format(f_path))
logger.info('Start processing.')
predictions = proc.get_predictions(self._sample_rate,
self._process_buf)
logger.info('Predictions: {}'.format(
format_predictions(predictions)))
logger.info('Stop processing.')
self._process_buf = None
self._ask_data.set()
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data.set()
try:
producer = KafkaProducer(
bootstrap_servers=config['KAFKA']['bootstrap_servers'])
except NoBrokersAvailable:
producer = None
print("No brokers available, running off-line")
while True:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data.clear()
if self._save_path:
f_path = os.path.join(
self._save_path,
'record_{:.0f}.wav'.format(time.time()))
wavfile.write(f_path, self._sample_rate, self._process_buf)
predictions = proc.get_predictions(self._sample_rate,
self._process_buf)
for prediction in predictions:
if prediction[1] > 0.4:
if producer is not None:
producer.send(config['KAFKA']['topic'],
bytes(str(prediction[0]).encode()))
print(prediction[0])
self._process_buf = None
self._ask_data.set()
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data.set()
while True:
i = 0
pred_values = [
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0
]
while i < 50:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data.clear()
if self._save_path:
f_path = os.path.join(
self._save_path,
'record_{:.0f}.wav'.format(time.time()))
wavfile.write(f_path, self._sample_rate,
self._process_buf)
logger.info('"{}" saved.'.format(f_path))
logger.info(str(i) + ' Start processing.')
predictions = proc.get_predictions(self._sample_rate,
self._process_buf)
for p in predictions:
pred_values = [
x + y
for x, y in zip(pred_values, get_labels(p[0]))
]
# print(get_labels(x[0]))
# logger.info(
# 'Predictions: {}'.format(format_predictions(predictions))
# )
logger.info('Stop processing.')
self._process_buf = None
self._ask_data.set()
i += 1
#arr_index = np.argmax(np.array(pred_values))
arr_index = np.argwhere(pred_values == np.amax(pred_values))
arr_index = arr_index.flatten().tolist()
#print(pred_values[int(arr_index)])
#print(arr_index)
# if(isinstance(arr_index,np.int64)):
# print(labels[int(arr_index)])
# else:
for x in arr_index:
print(labels[int(x)])
print(pred_values)
def process_file(wav_file):
sr, data = wavfile.read(wav_file)
if data.dtype != np.int16:
raise TypeError('Bad sample type: %r' % data.dtype)
# local import to reduce start-up time
from audio.processor import WavProcessor, format_predictions
with WavProcessor() as proc:
predictions = proc.get_predictions(sr, data)
print(format_predictions(predictions))
def gain_features(wav_root='Audio_Dataset',
save_path='./EmotiW_youtubeAudio.npz'):
# save path
data_path = []
log_spectrograms_save = []
embeddings_save = []
quantized_embeddings_save = []
proc = WavProcessor()
for wav_path in tqdm.tqdm(glob.glob(wav_root + '/*')):
wav_name = os.path.basename(wav_path).split('.')[0]
sr, data = wavfile.read(
wav_path
) # sr: sample rate; data: original sample data; data.type uint16
if data.dtype != np.int16:
raise TypeError('Bad sample type: %r' % data.dtype)
if len(data) < sr: continue # less than 1s
examples_batch, features, quantized_embeddings, predictions = proc.get_predictions(
sr, data)
# examples_batch: [3, 96, 60]
# features: [3, 128]
# quantized_embeddings: [3, 128]
frame_num = np.shape(examples_batch)[0]
data_path.append({'pic_path': wav_name})
log_spectrograms_save.append(examples_batch.reshape(frame_num, -1))
embeddings_save.append(features)
quantized_embeddings_save.append(quantized_embeddings)
np.savez_compressed(
save_path,
data_path=data_path,
logspec=log_spectrograms_save,
emb=embeddings_save,
qemb=quantized_embeddings_save,
)
def process_dir(start_dir, out_file):
f = open(out_file, 'w')
results = []
with WavProcessor() as proc:
for dirpath, dirs, files in os.walk(start_dir):
for filename in [f for f in files if f.lower().endswith('.wav')]:
fname = os.path.join(dirpath, filename)
result = process_file(fname, proc)
results.append((fname, tuple(result)))
print('%s %s' % (fname, result))
print(json.dumps(results), file=f)
f.close()
def process_file(wav_file, ckpt_file, labels_file):
sr, data = wavfile.read(wav_file)
print(wav_file, end=' ')
if data.dtype != np.int16:
raise TypeError('Bad sample type: %r' % data.dtype)
predictions = []
with WavProcessor(ckpt_file, labels_file) as proc:
try:
predictions = proc.get_predictions(sr, data)
print(format_predictions(predictions))
except:
print('CORRUPTED!')
return predictions
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data.set()
while True:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data.clear()
if self._save_path:
f_path = os.path.join(
self._save_path,
'record_{:.0f}.wav'.format(time.time()))
wavfile.write(f_path, self._sample_rate, self._process_buf)
logger.info('"{}" saved.'.format(f_path))
logger.info('Start processing.')
predictions = proc.get_predictions(self._sample_rate,
self._process_buf)
logger.info('Predictions: {}'.format(
format_predictions(predictions)))
my_dict = {x[0]: x[1] for x in predictions}
msg_json = {
"app_id": "openears",
"dev_id": "TEST001",
"payload_fields": my_dict,
"time": int(time.time() * 1e3)
}
msg_str = json.dumps(msg_json)
auth = {"username": mqtt_user, "password": mqtt_password}
publish.single("pipeline/openears/TEST001",
payload=msg_str,
hostname=mqtt_host,
port=mqtt_port,
auth=auth)
print('message sent')
logger.info('Stop processing.')
self._process_buf = None
self._ask_data.set()
def _process_loop(self):
with WavProcessor() as proc:
self._ask_data.set()
while True:
if self._process_buf is None:
# Waiting for data to process
time.sleep(self._processor_sleep_time)
continue
self._ask_data.clear()
if self._save_path:
f_path = os.path.join(
self._save_path,
'record_{:.0f}.wav'.format(time.time()))
wavfile.write(f_path, self._sample_rate, self._process_buf)
logger.info('"{}" saved.'.format(f_path))
logger.info('Start processing.')
predictions = proc.get_predictions(self._sample_rate,
self._process_buf)
for i in predictions:
if i[0] == "Vehicle" and (float(i[1]) > .2):
print("OBS Release")
GPIO.output(pin, GPIO.HIGH)
else:
GPIO.output(pin, GPIO.LOW)
print(i[0])
print(i[1])
#print("Type: ")
#print(type(predictions))
#logger.info(
#'Predictions: {}'.format(format_predictions(predictions))
#)
logger.info('Stop processing.')
self._process_buf = None
self._ask_data.set()
from pydub import AudioSegment
from audio.processor import WavProcessor, format_predictions
df = pd.read_csv('labels.csv')
df = df.set_index('index')
# print(df)
sig = AudioSegment.from_file('kitchen.wav', format="wav")
_start = 0
_end = 6
for x in range(0, 100):
start = int(_start * 1000)
end = int(_end * 1000)
segment = sig[start:end]
samples = np.array(segment.get_array_of_samples())
new_sig = samples.astype(np.float32)
with WavProcessor() as proc:
predictions = proc.get_predictions(16000, new_sig)
for p in predictions:
print(str(_start / 60), ":", str(_end / 60), " - ",
df.loc[p[0]].values[1], p[1])
_start += 6
_end += 6
import math
import operator
import json
from collections import Counter
import numpy as np
import pandas as pd
from scipy.io import wavfile
from pydub import AudioSegment
from audio.processor import WavProcessor, format_predictions
from log_config import LOGGING
proc = WavProcessor()
def get_labels(csv_file='labels.csv'):
df = pd.read_csv(csv_file)
df = df.set_index('index')
return df
def get_label_val(idx):
return [0.0 if val != 1 else 1.0 for val in df.loc[idx].values]
def label_sig(sig, df=get_labels()):
global proc
