Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import requests
|
2 |
+
import base64
|
3 |
+
import json
|
4 |
+
import streamlit as st
|
5 |
+
from speechlib import Transcriptor
|
6 |
+
|
7 |
+
def transcribe_audio(file, log_folder, language, modelSize, ACCESS_TOKEN, voices_folder, quantization):
|
8 |
+
transcriptor = Transcriptor(file, log_folder, language, modelSize, ACCESS_TOKEN, voices_folder, quantization)
|
9 |
+
return transcriptor.whisper()
|
10 |
+
|
11 |
+
def transform_transcript(transcript):
|
12 |
+
result = []
|
13 |
+
for segment in transcript:
|
14 |
+
start_time, end_time, text, speaker = segment
|
15 |
+
result.append(f"{speaker} ({start_time:.1f} : {end_time:.1f}) : {text}")
|
16 |
+
return '\n'.join(result)
|
17 |
+
|
18 |
+
st.title('Audio Transcription App')
|
19 |
+
|
20 |
+
ACCESS_TOKEN = st.secrets["HF_TOKEN"]
|
21 |
+
|
22 |
+
uploaded_file = st.file_uploader("Загрузите аудиофайл", type=["mp4", "wav", "m4a"])
|
23 |
+
|
24 |
+
if uploaded_file is not None:
|
25 |
+
file_extension = uploaded_file.name.split(".")[-1] # Получаем расширение файла
|
26 |
+
temp_file_path = f"temp_file.{file_extension}" # Создаем временное имя файла с правильным расширением
|
27 |
+
|
28 |
+
with open(temp_file_path, "wb") as f:
|
29 |
+
f.write(uploaded_file.getbuffer())
|
30 |
+
|
31 |
+
log_folder = "logs"
|
32 |
+
language = "ru"
|
33 |
+
modelSize = "large"
|
34 |
+
voices_folder = ""
|
35 |
+
quantization = False
|
36 |
+
|
37 |
+
with st.spinner('Транскрибируем...'):
|
38 |
+
result = transcribe_audio(temp_file_path, log_folder, language, modelSize, ACCESS_TOKEN, voices_folder, quantization)
|
39 |
+
|
40 |
+
st.write("Результат транскрибации:")
|
41 |
+
transcript = transform_transcript(result)
|
42 |
+
st.text(transcript)
|
43 |
+
|
44 |
+
with st.spinner('Резюмируем...'):
|
45 |
+
username = st.secrets["GIGA_USERNAME"]
|
46 |
+
password = st.secrets["GIGA_SECRET"]
|
47 |
+
|
48 |
+
# Получаем строку с базовой авторизацией в формате Base64
|
49 |
+
auth_str = f'{username}:{password}'
|
50 |
+
auth_bytes = auth_str.encode('utf-8')
|
51 |
+
auth_base64 = base64.b64encode(auth_bytes).decode('utf-8')
|
52 |
+
url = 'https://ngw.devices.sberbank.ru:9443/api/v2/oauth'
|
53 |
+
|
54 |
+
headers = {
|
55 |
+
'Authorization': f'Basic {auth_base64}', # вставляем базовую авторизацию
|
56 |
+
'RqUID': '6f0b1291-c7f3-43c6-bb2e-9f3efb2dc98f',
|
57 |
+
'Content-Type': 'application/x-www-form-urlencoded',
|
58 |
+
'Accept': 'application/json'
|
59 |
+
}
|
60 |
+
|
61 |
+
data = {
|
62 |
+
'scope': 'GIGACHAT_API_PERS'
|
63 |
+
}
|
64 |
+
|
65 |
+
response = requests.post(url, headers=headers, data=data, verify=False)
|
66 |
+
access_token = response.json()['access_token']
|
67 |
+
print('Got access token')
|
68 |
+
|
69 |
+
url_completion = "https://gigachat.devices.sberbank.ru/api/v1/chat/completions"
|
70 |
+
|
71 |
+
data_copm = json.dumps({
|
72 |
+
"model": "GigaChat",
|
73 |
+
"messages": [
|
74 |
+
{
|
75 |
+
"role": "user",
|
76 |
+
"content": "Напиши резюме транскрибации звонка, текст которого приложен в ниже. Выдели самостоятельно цель встречи, потом описать ключевые моменты всей встречи. Потом выделить отдельные темы звонка и выделить ключевые моменты в них. Напиши итоги того, о чем договорились говорящие, если такое возможно выделить из текста. Транскрибация: " + transcript
|
77 |
+
}
|
78 |
+
],
|
79 |
+
"stream": False,
|
80 |
+
"max_tokens": 1024,
|
81 |
+
})
|
82 |
+
|
83 |
+
headers_comp = {
|
84 |
+
'Content-Type': 'application/json',
|
85 |
+
'Accept': 'application/json',
|
86 |
+
'Authorization': 'Bearer ' + access_token
|
87 |
+
}
|
88 |
+
|
89 |
+
response = requests.post(url_completion, headers=headers_comp, data=data_copm, verify=False)
|
90 |
+
response_data = response.json()
|
91 |
+
answer_from_llm = response_data['choices'][0]['message']['content']
|
92 |
+
|
93 |
+
st.write("Результат резюмирования:")
|
94 |
+
st.text(answer_from_llm)
|