-
Notifications
You must be signed in to change notification settings - Fork 124
/
server.py
151 lines (133 loc) · 5.41 KB
/
server.py
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
import base64
import json
import io
import os
import re
import threading
import wave
import zlib
import cherrypy
import numpy as np
from service import Caffe2LabelService, TorchLabelService, TrainingService
from service import stride
def json_in(f):
def merge_dicts(x, y):
z = x.copy()
z.update(y)
return z
def wrapper(*args, **kwargs):
cl = cherrypy.request.headers["Content-Length"]
data = json.loads(cherrypy.request.body.read(int(cl)).decode("utf-8"))
kwargs = merge_dicts(kwargs, data)
return f(*args, **kwargs)
return wrapper
class TrainEndpoint(object):
exposed = True
def __init__(self, train_service, label_service):
self.train_service = train_service
self.label_service = label_service
@cherrypy.tools.json_out()
def POST(self):
return dict(success=self.train_service.run_train_script(callback=self.label_service.reload))
@cherrypy.tools.json_out()
def GET(self):
return dict(in_progress=self.train_service.script_running)
class DataEndpoint(object):
exposed = True
def __init__(self, train_service):
self.train_service = train_service
@cherrypy.tools.json_out()
@json_in
def POST(self, **kwargs):
wav_data = zlib.decompress(base64.b64decode(kwargs["wav_data"]))
positive = kwargs["positive"]
for _ in range(3):
self.train_service.write_example(wav_data, positive=positive)
success = dict(success=True)
if not positive:
return success
neg_examples = self.train_service.generate_contrastive(wav_data)
if not neg_examples:
return success
for example in neg_examples:
self.train_service.write_example(example.byte_data, positive=False, tag="gen")
return success
@cherrypy.tools.json_out()
def DELETE(self):
self.train_service.clear_examples(positive=True)
self.train_service.clear_examples(positive=False, tag="gen")
return dict(success=True)
class EvaluateEndpoint(object):
exposed = True
def __init__(self, label_service):
self.label_service = label_service
@cherrypy.tools.json_out()
@json_in
def POST(self, **kwargs):
folders = kwargs["folders"]
indices = kwargs.get("indices", [])
accuracy = self.label_service.evaluate(folders, indices)
return dict(accuracy=accuracy)
class ListenEndpoint(object):
exposed = True
def __init__(self, label_service, stride_size=500, min_keyword_prob=0.85, keyword="command"):
"""The REST API endpoint that determines if audio contains the keyword.
Args:
label_service: The labelling service to use
stride_size: The stride in milliseconds of the 1-second window to use. It should divide 1000 ms.
min_keyword_prob: The minimum probability the keyword must take in order to be classified as such
keyword: The keyword
"""
self.label_service = label_service
self.stride_size = stride_size
self.min_keyword_prob = min_keyword_prob
self.keyword = keyword
@cherrypy.tools.json_out()
@json_in
def POST(self, **kwargs):
wav_data = zlib.decompress(base64.b64decode(kwargs["wav_data"]))
labels = {}
for data in stride(wav_data, int(2 * 16000 * self.stride_size / 1000), 2 * 16000):
label, prob = self.label_service.label(data)
try:
labels[label] += float(prob)
except KeyError:
labels[label] = float(prob)
if label == "command" and prob >= self.min_keyword_prob and kwargs["method"] == "command_tagging":
return dict(contains_command=True)
return dict(contains_command=False) if kwargs["method"] == "command_tagging" else labels
def make_abspath(rel_path):
if not os.path.isabs(rel_path):
rel_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), rel_path)
return rel_path
def load_service(config):
model_path = make_abspath(config["model_path"])
commands = ["__silence__", "__unknown__"]
commands.extend(config["commands"].split(","))
backend = config["backend"]
if backend.lower() == "caffe2":
lbl_service = Caffe2LabelService(model_path, commands)
elif backend.lower() == "pytorch":
lbl_service = TorchLabelService(model_path, labels=commands, no_cuda=config["model_options"]["no_cuda"])
else:
raise ValueError("Backend {} not supported!".format(backend))
return lbl_service
def start(config):
cherrypy.config.update({
"environment": "production",
"log.screen": True
})
cherrypy.config.update(config["server"])
rest_config = {"/": {
"request.dispatch": cherrypy.dispatch.MethodDispatcher()
}}
train_script = make_abspath(config["train_script"])
speech_dataset_path = make_abspath(config["speech_dataset_path"])
lbl_service = load_service(config)
train_service = TrainingService(train_script, speech_dataset_path, config["model_options"])
cherrypy.tree.mount(ListenEndpoint(lbl_service), "/listen", rest_config)
cherrypy.tree.mount(DataEndpoint(train_service), "/data", rest_config)
cherrypy.tree.mount(EvaluateEndpoint(lbl_service), "/evaluate", rest_config)
cherrypy.tree.mount(TrainEndpoint(train_service, lbl_service), "/train", rest_config)
cherrypy.engine.start()
cherrypy.engine.block()