Inicio Explorar Ayuda
Registro Iniciar sesión
J1X
/
deepspeech-server
Vigilar 2
Destacar 0
Fork 0

Explorar el Código
Initial commit

James Peret %!s(int64=5) %!d(string=hace) años
commit
8e44e65008
Se han modificado 2 ficheros con 168 adiciones y 0 borrados
Mostrar Estadísticas de Diff
  1. 7 0
      .gitignore
  2. 161 0
      deepspeech-server.py

+ 7 - 0
.gitignore
Ver Fichero 

@@ -0,0 +1,7 @@
1 
+.DS_Store
2 
+*/.DS_Store
3 
+audio-0.6.0.tar.gz
4 
+audio/*
5 
+deepspeech-0.6.0-models.tar.gz
6 
+deepspeech-0.6.0-models/*
7 
+temp_audio/*

+ 161 - 0
deepspeech-server.py
Ver Fichero 

@@ -0,0 +1,161 @@
1 
+#!/usr/bin/env python
2 
+# -*- coding: utf-8 -*-
3 
+from __future__ import absolute_import, division, print_function
4 
+
5 
+import argparse
6 
+import numpy as np
7 
+import shlex
8 
+import subprocess
9 
+import sys
10 
+import wave
11 
+import json
12 
+
13 
+import os
14 
+from bottle import route, run, template, request, response
15 
+
16 
+from deepspeech import Model, printVersions
17 
+from timeit import default_timer as timer
18 
+
19 
+try:
20 
+    from shhlex import quote
21 
+except ImportError:
22 
+    from pipes import quote
23 
+
24 
+model = "deepspeech-0.6.0-models/output_graph.pbmm"
25 
+lm = "deepspeech-0.6.0-models/lm.binary"
26 
+trie = "deepspeech-0.6.0-models/trie"
27 
+beam_width = 500
28 
+lm_alpha = 0.75
29 
+lm_beta = 1.85
30 
+desired_sample_rate = 16000
31 
+ds = {}
32 
+
33 
+def convert_samplerate(audio_path, desired_sample_rate):
34 
+    sox_cmd = 'sox {} --type raw --bits 16 --channels 1 --rate {} --encoding signed-integer --endian little --compression 0.0 --no-dither - '.format(quote(audio_path), desired_sample_rate)
35 
+    try:
36 
+        output = subprocess.check_output(shlex.split(sox_cmd), stderr=subprocess.PIPE)
37 
+    except subprocess.CalledProcessError as e:
38 
+        raise RuntimeError('SoX returned non-zero status: {}'.format(e.stderr))
39 
+    except OSError as e:
40 
+        raise OSError(e.errno, 'SoX not found, use {}hz files or install it: {}'.format(desired_sample_rate, e.strerror))
41 
+
42 
+    return desired_sample_rate, np.frombuffer(output, np.int16)
43 
+
44 
+
45 
+def metadata_to_string(metadata):
46 
+    return ''.join(item.character for item in metadata.items)
47 
+
48 
+def words_from_metadata(metadata):
49 
+    word = ""
50 
+    word_list = []
51 
+    word_start_time = 0
52 
+    # Loop through each character
53 
+    for i in range(0, metadata.num_items):
54 
+        item = metadata.items[i]
55 
+        # Append character to word if it's not a space
56 
+        if item.character != " ":
57 
+            word = word + item.character
58 
+        # Word boundary is either a space or the last character in the array
59 
+        if item.character == " " or i == metadata.num_items - 1:
60 
+            word_duration = item.start_time - word_start_time
61 
+
62 
+            if word_duration < 0:
63 
+                word_duration = 0
64 
+
65 
+            each_word = dict()
66 
+            each_word["word"] = word
67 
+            each_word["start_time "] = round(word_start_time, 4)
68 
+            each_word["duration"] = round(word_duration, 4)
69 
+
70 
+            word_list.append(each_word)
71 
+            # Reset
72 
+            word = ""
73 
+            word_start_time = 0
74 
+        else:
75 
+            if len(word) == 1:
76 
+                # Log the start time of the new word
77 
+                word_start_time = item.start_time
78 
+
79 
+    return word_list
80 
+
81 
+
82 
+def metadata_json_output(metadata):
83 
+    json_result = dict()
84 
+    json_result["words"] = words_from_metadata(metadata)
85 
+    json_result["confidence"] = metadata.confidence
86 
+    return json.dumps(json_result)
87 
+
88 
+
89 
+
90 
+class VersionAction(argparse.Action):
91 
+    def __init__(self, *args, **kwargs):
92 
+        super(VersionAction, self).__init__(nargs=0, *args, **kwargs)
93 
+
94 
+    def __call__(self, *args, **kwargs):
95 
+        printVersions()
96 
+        exit(0)
97 
+
98 
+@route('/upload', method='POST')
99 
+def do_upload():
100 
+    print('received a request')
101 
+    response.content_type = 'application/json'
102 
+    upload = request.POST['file']
103 
+    name, ext = os.path.splitext(upload.filename)
104 
+    if ext not in ('.wav'):
105 
+        return "File extension not allowed."
106 
+
107 
+    save_path = "temp_audio"
108 
+    if not os.path.exists(save_path):
109 
+        os.makedirs(save_path)
110 
+
111 
+    file_path = "{path}/{file}".format(path=save_path, file=upload.filename)
112 
+    print(file_path)
113 
+    upload.save(file_path, overwrite=True)
114 
+    data = process(file_path)
115 
+    return data
116 
+    #return { "message": "File successfully saved to '{0}'.".format(save_path)}
117 
+
118 
+def start():
119 
+    global ds
120 
+    print('Loading model from file {}'.format(model), file=sys.stderr)
121 
+    model_load_start = timer()
122 
+    ds = Model(model, beam_width)
123 
+    model_load_end = timer() - model_load_start
124 
+    print('Loaded model in {:.3}s.'.format(model_load_end), file=sys.stderr)
125 
+
126 
+    #desired_sample_rate = ds.sampleRate()
127 
+
128 
+    if lm and trie:
129 
+        print('Loading language model from files {} {}'.format(lm, trie), file=sys.stderr)
130 
+        lm_load_start = timer()
131 
+        ds.enableDecoderWithLM(lm, trie, lm_alpha, lm_beta)
132 
+        lm_load_end = timer() - lm_load_start
133 
+        print('Loaded language model in {:.3}s.'.format(lm_load_end), file=sys.stderr)
134 
+
135 
+def process(audio):
136 
+    global ds
137 
+    fin = wave.open(audio, 'rb')
138 
+    fs = fin.getframerate()
139 
+    if fs != desired_sample_rate:
140 
+        print('Warning: original sample rate ({}) is different than {}hz. Resampling might produce erratic speech recognition.'.format(fs, desired_sample_rate), file=sys.stderr)
141 
+        fs, audio = convert_samplerate(audio, desired_sample_rate)
142 
+    else:
143 
+        audio = np.frombuffer(fin.readframes(fin.getnframes()), np.int16)
144 
+
145 
+    audio_length = fin.getnframes() * (1/fs)
146 
+    fin.close()
147 
+
148 
+    print('Running inference.', file=sys.stderr)
149 
+    inference_start = timer()
150 
+    data = metadata_json_output(ds.sttWithMetadata(audio))
151 
+    inference_end = timer() - inference_start
152 
+    print('Inference took %0.3fs for %0.3fs audio file.' % (inference_end, audio_length), file=sys.stderr)
153 
+    print(data)
154 
+    return data
155 
+
156 
+if __name__ == '__main__':
157 
+    start()
158 
+    run(host='localhost', port=8050)
159 
+
160 
+if __name__ == '__main__':
161 
+    main()