Spaces:
Build error
Build error
File size: 7,355 Bytes
98f685a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 |
import os
import sys
sys.path.insert(1, os.path.join(sys.path[0], '../utils'))
import numpy as np
import argparse
import librosa
import matplotlib.pyplot as plt
import torch
from utilities import create_folder, get_filename
from models import *
from pytorch_utils import move_data_to_device
import config
def audio_tagging(args):
"""Inference audio tagging result of an audio clip.
"""
# Arugments & parameters
sample_rate = args.sample_rate
window_size = args.window_size
hop_size = args.hop_size
mel_bins = args.mel_bins
fmin = args.fmin
fmax = args.fmax
model_type = args.model_type
checkpoint_path = args.checkpoint_path
audio_path = args.audio_path
device = torch.device('cuda') if args.cuda and torch.cuda.is_available() else torch.device('cpu')
classes_num = config.classes_num
labels = config.labels
# Model
Model = eval(model_type)
model = Model(sample_rate=sample_rate, window_size=window_size,
hop_size=hop_size, mel_bins=mel_bins, fmin=fmin, fmax=fmax,
classes_num=classes_num)
checkpoint = torch.load(checkpoint_path, map_location=device)
model.load_state_dict(checkpoint['model'])
# Parallel
if 'cuda' in str(device):
model.to(device)
print('GPU number: {}'.format(torch.cuda.device_count()))
model = torch.nn.DataParallel(model)
else:
print('Using CPU.')
# Load audio
(waveform, _) = librosa.core.load(audio_path, sr=sample_rate, mono=True)
waveform = waveform[None, :] # (1, audio_length)
waveform = move_data_to_device(waveform, device)
# Forward
with torch.no_grad():
model.eval()
batch_output_dict = model(waveform, None)
clipwise_output = batch_output_dict['clipwise_output'].data.cpu().numpy()[0]
"""(classes_num,)"""
sorted_indexes = np.argsort(clipwise_output)[::-1]
# Print audio tagging top probabilities
for k in range(10):
print('{}: {:.3f}'.format(np.array(labels)[sorted_indexes[k]],
clipwise_output[sorted_indexes[k]]))
# Print embedding
if 'embedding' in batch_output_dict.keys():
embedding = batch_output_dict['embedding'].data.cpu().numpy()[0]
print('embedding: {}'.format(embedding.shape))
return clipwise_output, labels
def sound_event_detection(args):
"""Inference sound event detection result of an audio clip.
"""
# Arugments & parameters
sample_rate = args.sample_rate
window_size = args.window_size
hop_size = args.hop_size
mel_bins = args.mel_bins
fmin = args.fmin
fmax = args.fmax
model_type = args.model_type
checkpoint_path = args.checkpoint_path
audio_path = args.audio_path
device = torch.device('cuda') if args.cuda and torch.cuda.is_available() else torch.device('cpu')
classes_num = config.classes_num
labels = config.labels
frames_per_second = sample_rate // hop_size
# Paths
fig_path = os.path.join('results', '{}.png'.format(get_filename(audio_path)))
create_folder(os.path.dirname(fig_path))
# Model
Model = eval(model_type)
model = Model(sample_rate=sample_rate, window_size=window_size,
hop_size=hop_size, mel_bins=mel_bins, fmin=fmin, fmax=fmax,
classes_num=classes_num)
checkpoint = torch.load(checkpoint_path, map_location=device)
model.load_state_dict(checkpoint['model'])
# Parallel
print('GPU number: {}'.format(torch.cuda.device_count()))
model = torch.nn.DataParallel(model)
if 'cuda' in str(device):
model.to(device)
# Load audio
(waveform, _) = librosa.core.load(audio_path, sr=sample_rate, mono=True)
waveform = waveform[None, :] # (1, audio_length)
waveform = move_data_to_device(waveform, device)
# Forward
with torch.no_grad():
model.eval()
batch_output_dict = model(waveform, None)
framewise_output = batch_output_dict['framewise_output'].data.cpu().numpy()[0]
"""(time_steps, classes_num)"""
print('Sound event detection result (time_steps x classes_num): {}'.format(
framewise_output.shape))
sorted_indexes = np.argsort(np.max(framewise_output, axis=0))[::-1]
top_k = 10 # Show top results
top_result_mat = framewise_output[:, sorted_indexes[0 : top_k]]
"""(time_steps, top_k)"""
# Plot result
stft = librosa.core.stft(y=waveform[0].data.cpu().numpy(), n_fft=window_size,
hop_length=hop_size, window='hann', center=True)
frames_num = stft.shape[-1]
fig, axs = plt.subplots(2, 1, sharex=True, figsize=(10, 4))
axs[0].matshow(np.log(np.abs(stft)), origin='lower', aspect='auto', cmap='jet')
axs[0].set_ylabel('Frequency bins')
axs[0].set_title('Log spectrogram')
axs[1].matshow(top_result_mat.T, origin='upper', aspect='auto', cmap='jet', vmin=0, vmax=1)
axs[1].xaxis.set_ticks(np.arange(0, frames_num, frames_per_second))
axs[1].xaxis.set_ticklabels(np.arange(0, frames_num / frames_per_second))
axs[1].yaxis.set_ticks(np.arange(0, top_k))
axs[1].yaxis.set_ticklabels(np.array(labels)[sorted_indexes[0 : top_k]])
axs[1].yaxis.grid(color='k', linestyle='solid', linewidth=0.3, alpha=0.3)
axs[1].set_xlabel('Seconds')
axs[1].xaxis.set_ticks_position('bottom')
plt.tight_layout()
plt.savefig(fig_path)
print('Save sound event detection visualization to {}'.format(fig_path))
return framewise_output, labels
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Example of parser. ')
subparsers = parser.add_subparsers(dest='mode')
parser_at = subparsers.add_parser('audio_tagging')
parser_at.add_argument('--sample_rate', type=int, default=32000)
parser_at.add_argument('--window_size', type=int, default=1024)
parser_at.add_argument('--hop_size', type=int, default=320)
parser_at.add_argument('--mel_bins', type=int, default=64)
parser_at.add_argument('--fmin', type=int, default=50)
parser_at.add_argument('--fmax', type=int, default=14000)
parser_at.add_argument('--model_type', type=str, required=True)
parser_at.add_argument('--checkpoint_path', type=str, required=True)
parser_at.add_argument('--audio_path', type=str, required=True)
parser_at.add_argument('--cuda', action='store_true', default=False)
parser_sed = subparsers.add_parser('sound_event_detection')
parser_sed.add_argument('--sample_rate', type=int, default=32000)
parser_sed.add_argument('--window_size', type=int, default=1024)
parser_sed.add_argument('--hop_size', type=int, default=320)
parser_sed.add_argument('--mel_bins', type=int, default=64)
parser_sed.add_argument('--fmin', type=int, default=50)
parser_sed.add_argument('--fmax', type=int, default=14000)
parser_sed.add_argument('--model_type', type=str, required=True)
parser_sed.add_argument('--checkpoint_path', type=str, required=True)
parser_sed.add_argument('--audio_path', type=str, required=True)
parser_sed.add_argument('--cuda', action='store_true', default=False)
args = parser.parse_args()
if args.mode == 'audio_tagging':
audio_tagging(args)
elif args.mode == 'sound_event_detection':
sound_event_detection(args)
else:
raise Exception('Error argument!') |