Spaces:
Sleeping
Sleeping
Create utils.py
Browse files
utils.py
ADDED
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Some utility functions for the app."""
|
2 |
+
from base64 import b64encode
|
3 |
+
from io import BytesIO
|
4 |
+
|
5 |
+
from gtts import gTTS
|
6 |
+
from mtranslate import translate
|
7 |
+
from speech_recognition import AudioFile, Recognizer
|
8 |
+
from transformers import (BlenderbotSmallForConditionalGeneration,
|
9 |
+
BlenderbotSmallTokenizer)
|
10 |
+
|
11 |
+
|
12 |
+
def stt(audio: object, language: str) -> str:
|
13 |
+
"""Converts speech to text.
|
14 |
+
Args:
|
15 |
+
audio: record of user speech
|
16 |
+
Returns:
|
17 |
+
text (str): recognized speech of user
|
18 |
+
"""
|
19 |
+
|
20 |
+
# Create a Recognizer object
|
21 |
+
r = Recognizer()
|
22 |
+
# Open the audio file
|
23 |
+
with AudioFile(audio) as source:
|
24 |
+
# Listen for the data (load audio to memory)
|
25 |
+
audio_data = r.record(source)
|
26 |
+
# Transcribe the audio using Google's speech-to-text API
|
27 |
+
text = r.recognize_google(audio_data, language=language)
|
28 |
+
return text
|
29 |
+
|
30 |
+
|
31 |
+
def to_en_translation(text: str, language: str) -> str:
|
32 |
+
"""Translates text from specified language to English.
|
33 |
+
Args:
|
34 |
+
text (str): input text
|
35 |
+
language (str): desired language
|
36 |
+
Returns:
|
37 |
+
str: translated text
|
38 |
+
"""
|
39 |
+
return translate(text, "en", language)
|
40 |
+
|
41 |
+
|
42 |
+
def from_en_translation(text: str, language: str) -> str:
|
43 |
+
"""Translates text from english to specified language.
|
44 |
+
Args:
|
45 |
+
text (str): input text
|
46 |
+
language (str): desired language
|
47 |
+
Returns:
|
48 |
+
str: translated text
|
49 |
+
"""
|
50 |
+
return translate(text, language, "en")
|
51 |
+
|
52 |
+
|
53 |
+
class TextGenerationPipeline:
|
54 |
+
"""Pipeline for text generation of blenderbot model.
|
55 |
+
Returns:
|
56 |
+
str: generated text
|
57 |
+
"""
|
58 |
+
|
59 |
+
# load tokenizer and the model
|
60 |
+
model_name = "facebook/blenderbot_small-90M"
|
61 |
+
tokenizer = BlenderbotSmallTokenizer.from_pretrained(model_name)
|
62 |
+
model = BlenderbotSmallForConditionalGeneration.from_pretrained(model_name)
|
63 |
+
|
64 |
+
def __init__(self, **kwargs):
|
65 |
+
"""Specififying text generation parameters.
|
66 |
+
For example: max_length=100 which generates text shorter than
|
67 |
+
100 tokens. Visit:
|
68 |
+
https://huggingface.co/docs/transformers/main_classes/text_generation
|
69 |
+
for more parameters
|
70 |
+
"""
|
71 |
+
self.__dict__.update(kwargs)
|
72 |
+
|
73 |
+
def preprocess(self, text) -> str:
|
74 |
+
"""Tokenizes input text.
|
75 |
+
Args:
|
76 |
+
text (str): user specified text
|
77 |
+
Returns:
|
78 |
+
torch.Tensor (obj): text representation as tensors
|
79 |
+
"""
|
80 |
+
return self.tokenizer(text, return_tensors="pt")
|
81 |
+
|
82 |
+
def postprocess(self, outputs) -> str:
|
83 |
+
"""Converts tensors into text.
|
84 |
+
Args:
|
85 |
+
outputs (torch.Tensor obj): model text generation output
|
86 |
+
Returns:
|
87 |
+
str: generated text
|
88 |
+
"""
|
89 |
+
return self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
90 |
+
|
91 |
+
def __call__(self, text: str) -> str:
|
92 |
+
"""Generates text from input text.
|
93 |
+
Args:
|
94 |
+
text (str): user specified text
|
95 |
+
Returns:
|
96 |
+
str: generated text
|
97 |
+
"""
|
98 |
+
tokenized_text = self.preprocess(text)
|
99 |
+
output = self.model.generate(**tokenized_text, **self.__dict__)
|
100 |
+
return self.postprocess(output)
|
101 |
+
|
102 |
+
|
103 |
+
def tts(text: str, language: str) -> object:
|
104 |
+
"""Converts text into audio object.
|
105 |
+
Args:
|
106 |
+
text (str): generated answer of bot
|
107 |
+
Returns:
|
108 |
+
object: text to speech object
|
109 |
+
"""
|
110 |
+
return gTTS(text=text, lang=language, slow=False)
|
111 |
+
|
112 |
+
|
113 |
+
def tts_to_bytesio(tts_object: object) -> bytes:
|
114 |
+
"""Converts tts object to bytes.
|
115 |
+
Args:
|
116 |
+
tts_object (object): audio object obtained from gtts
|
117 |
+
Returns:
|
118 |
+
bytes: audio bytes
|
119 |
+
"""
|
120 |
+
bytes_object = BytesIO()
|
121 |
+
tts_object.write_to_fp(bytes_object)
|
122 |
+
bytes_object.seek(0)
|
123 |
+
return bytes_object.getvalue()
|
124 |
+
|
125 |
+
|
126 |
+
def html_audio_autoplay(bytes: bytes) -> object:
|
127 |
+
"""Creates html object for autoplaying audio at gradio app.
|
128 |
+
Args:
|
129 |
+
bytes (bytes): audio bytes
|
130 |
+
Returns:
|
131 |
+
object: html object that provides audio autoplaying
|
132 |
+
"""
|
133 |
+
b64 = b64encode(bytes).decode()
|
134 |
+
html = f"""
|
135 |
+
<audio controls autoplay>
|
136 |
+
<source src="data:audio/wav;base64,{b64}" type="audio/wav">
|
137 |
+
</audio>
|
138 |
+
"""
|
139 |
+
return html
|