Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -4,24 +4,24 @@ import whisper
|
|
4 |
import tempfile
|
5 |
import gradio as gr
|
6 |
from pydub import AudioSegment
|
7 |
-
import fitz # PyMuPDF
|
8 |
-
import docx #
|
9 |
-
import pandas as pd #
|
10 |
-
# from google.colab import userdata #
|
11 |
import requests
|
12 |
from bs4 import BeautifulSoup
|
13 |
|
14 |
-
#
|
15 |
# openai.api_key = userdata.get('OPENAI_API_KEY')
|
16 |
|
17 |
-
#
|
18 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
19 |
|
20 |
-
#
|
21 |
model = whisper.load_model("large")
|
22 |
|
23 |
def preprocess_audio(audio_file):
|
24 |
-
"""
|
25 |
try:
|
26 |
audio = AudioSegment.from_file(audio_file)
|
27 |
audio = audio.apply_gain(-audio.dBFS + (-20))
|
@@ -29,150 +29,150 @@ def preprocess_audio(audio_file):
|
|
29 |
audio.export(temp_file.name, format="mp3")
|
30 |
return temp_file.name
|
31 |
except Exception as e:
|
32 |
-
return f"Error
|
33 |
|
34 |
-
def
|
35 |
-
"""Transcribe
|
36 |
try:
|
37 |
-
|
38 |
-
|
39 |
-
return
|
40 |
except Exception as e:
|
41 |
-
return f"Error
|
42 |
|
43 |
-
def
|
44 |
-
"""
|
45 |
try:
|
46 |
-
if
|
47 |
-
doc = fitz.open(
|
48 |
-
return "\n".join([
|
49 |
-
elif
|
50 |
-
doc = docx.Document(
|
51 |
-
return "\n".join([
|
52 |
-
elif
|
53 |
-
return pd.read_excel(
|
54 |
-
elif
|
55 |
-
return pd.read_csv(
|
56 |
else:
|
57 |
-
return "
|
58 |
except Exception as e:
|
59 |
-
return f"Error
|
60 |
|
61 |
-
def
|
62 |
-
"""
|
63 |
try:
|
64 |
response = requests.get(url)
|
65 |
response.raise_for_status()
|
66 |
soup = BeautifulSoup(response.content, 'html.parser')
|
67 |
return soup.get_text()
|
68 |
except Exception as e:
|
69 |
-
return f"Error
|
70 |
|
71 |
-
def
|
72 |
-
"""
|
73 |
-
|
74 |
-
num_audios = 5 * 3 # 5 audios * 3
|
75 |
audios = args[:num_audios]
|
76 |
-
|
77 |
|
78 |
for url in urls.split():
|
79 |
if url:
|
80 |
-
|
81 |
|
82 |
-
for
|
83 |
-
if
|
84 |
-
|
85 |
|
86 |
for i in range(0, len(audios), 3):
|
87 |
-
audio_file,
|
88 |
if audio_file is not None:
|
89 |
-
|
90 |
|
91 |
-
|
92 |
|
93 |
-
for idx, data in enumerate(
|
94 |
if data["audio"] is not None:
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
if
|
99 |
-
|
100 |
-
|
101 |
else:
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
-
|
111 |
-
-
|
112 |
-
-
|
113 |
-
-
|
114 |
-
-
|
115 |
-
-
|
116 |
-
-
|
117 |
"""
|
118 |
|
119 |
prompt = f"""
|
120 |
-
{
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
{
|
128 |
"""
|
129 |
|
130 |
try:
|
131 |
-
|
132 |
model="gpt-3.5-turbo",
|
133 |
messages=[{"role": "user", "content": prompt}],
|
134 |
temperature=0.1
|
135 |
)
|
136 |
-
|
137 |
-
return
|
138 |
except Exception as e:
|
139 |
-
return f"Error
|
140 |
|
141 |
with gr.Blocks() as demo:
|
142 |
-
gr.Markdown("##
|
143 |
with gr.Row():
|
144 |
with gr.Column(scale=2):
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
urls = gr.Textbox(label="URLs (
|
150 |
with gr.Column(scale=3):
|
151 |
-
inputs_list = [
|
152 |
with gr.Tabs():
|
153 |
for i in range(1, 6):
|
154 |
with gr.TabItem(f"Audio {i}"):
|
155 |
audio = gr.Audio(type="filepath", label=f"Audio {i}")
|
156 |
-
|
157 |
-
|
158 |
-
inputs_list.extend([audio,
|
159 |
for i in range(1, 6):
|
160 |
-
with gr.TabItem(f"
|
161 |
-
|
162 |
-
inputs_list.append(
|
163 |
|
164 |
-
gr.Markdown("---") #
|
165 |
|
166 |
with gr.Row():
|
167 |
-
|
168 |
|
169 |
-
gr.Markdown("---") #
|
170 |
|
171 |
with gr.Row():
|
172 |
-
|
173 |
with gr.Row():
|
174 |
-
|
175 |
|
176 |
-
|
177 |
|
178 |
-
demo.launch(share=True)
|
|
|
4 |
import tempfile
|
5 |
import gradio as gr
|
6 |
from pydub import AudioSegment
|
7 |
+
import fitz # PyMuPDF for handling PDFs
|
8 |
+
import docx # For handling .docx files
|
9 |
+
import pandas as pd # For handling .xlsx and .csv files
|
10 |
+
# from google.colab import userdata # Import userdata from google.colab
|
11 |
import requests
|
12 |
from bs4 import BeautifulSoup
|
13 |
|
14 |
+
# Configure your OpenAI API key using Google Colab userdata
|
15 |
# openai.api_key = userdata.get('OPENAI_API_KEY')
|
16 |
|
17 |
+
# Load environment variables from the Hugging Face environment
|
18 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
19 |
|
20 |
+
# Load the highest quality Whisper model once
|
21 |
model = whisper.load_model("large")
|
22 |
|
23 |
def preprocess_audio(audio_file):
|
24 |
+
"""Preprocess the audio file to improve quality."""
|
25 |
try:
|
26 |
audio = AudioSegment.from_file(audio_file)
|
27 |
audio = audio.apply_gain(-audio.dBFS + (-20))
|
|
|
29 |
audio.export(temp_file.name, format="mp3")
|
30 |
return temp_file.name
|
31 |
except Exception as e:
|
32 |
+
return f"Error preprocessing the audio file: {str(e)}"
|
33 |
|
34 |
+
def transcribe_audio(audio_file):
|
35 |
+
"""Transcribe an audio file."""
|
36 |
try:
|
37 |
+
file_path = preprocess_audio(audio_file) if isinstance(audio_file, str) else preprocess_audio(tempfile.NamedTemporaryFile(delete=False, suffix=".mp3", mode='w+b').name)
|
38 |
+
result = model.transcribe(file_path)
|
39 |
+
return result.get("text", "Error in transcription")
|
40 |
except Exception as e:
|
41 |
+
return f"Error processing the audio file: {str(e)}"
|
42 |
|
43 |
+
def read_document(document_path):
|
44 |
+
"""Read the content of a PDF, DOCX, XLSX or CSV document."""
|
45 |
try:
|
46 |
+
if document_path.endswith(".pdf"):
|
47 |
+
doc = fitz.open(document_path)
|
48 |
+
return "\n".join([page.get_text() for page in doc])
|
49 |
+
elif document_path.endswith(".docx"):
|
50 |
+
doc = docx.Document(document_path)
|
51 |
+
return "\n".join([paragraph.text for paragraph in doc.paragraphs])
|
52 |
+
elif document_path.endswith(".xlsx"):
|
53 |
+
return pd.read_excel(document_path).to_string()
|
54 |
+
elif document_path.endswith(".csv"):
|
55 |
+
return pd.read_csv(document_path).to_string()
|
56 |
else:
|
57 |
+
return "Unsupported file type. Please upload a PDF, DOCX, XLSX or CSV document."
|
58 |
except Exception as e:
|
59 |
+
return f"Error reading the document: {str(e)}"
|
60 |
|
61 |
+
def read_url(url):
|
62 |
+
"""Read the content of a URL."""
|
63 |
try:
|
64 |
response = requests.get(url)
|
65 |
response.raise_for_status()
|
66 |
soup = BeautifulSoup(response.content, 'html.parser')
|
67 |
return soup.get_text()
|
68 |
except Exception as e:
|
69 |
+
return f"Error reading the URL: {str(e)}"
|
70 |
|
71 |
+
def generate_news(instructions, facts, size, tone, urls, *args):
|
72 |
+
"""Generate a news article based on instructions, facts, URLs, documents, and transcriptions."""
|
73 |
+
knowledge_base = {"instructions": instructions, "facts": facts, "document_content": [], "audio_data": [], "url_content": []}
|
74 |
+
num_audios = 5 * 3 # 5 audios * 3 fields (audio, name, position)
|
75 |
audios = args[:num_audios]
|
76 |
+
documents = args[num_audios:]
|
77 |
|
78 |
for url in urls.split():
|
79 |
if url:
|
80 |
+
knowledge_base["url_content"].append(read_url(url))
|
81 |
|
82 |
+
for document in documents:
|
83 |
+
if document is not None:
|
84 |
+
knowledge_base["document_content"].append(read_document(document.name))
|
85 |
|
86 |
for i in range(0, len(audios), 3):
|
87 |
+
audio_file, name, position = audios[i:i+3]
|
88 |
if audio_file is not None:
|
89 |
+
knowledge_base["audio_data"].append({"audio": audio_file, "name": name, "position": position})
|
90 |
|
91 |
+
transcriptions_text, raw_transcriptions, total_direct_quotes = "", "", 0
|
92 |
|
93 |
+
for idx, data in enumerate(knowledge_base["audio_data"]):
|
94 |
if data["audio"] is not None:
|
95 |
+
transcription = transcribe_audio(data["audio"])
|
96 |
+
transcription_text = f'"{transcription}" - {data["name"]}, {data["position"]}'
|
97 |
+
raw_transcription = f'[Audio {idx + 1}]: "{transcription}" - {data["name"]}, {data["position"]}'
|
98 |
+
if total_direct_quotes < len(knowledge_base["audio_data"]) * 0.8:
|
99 |
+
transcriptions_text += transcription_text + "\n"
|
100 |
+
total_direct_quotes += 1
|
101 |
else:
|
102 |
+
transcriptions_text += f'{data["name"]} mentioned that {transcription}' + "\n"
|
103 |
+
raw_transcriptions += raw_transcription + "\n\n"
|
104 |
+
|
105 |
+
document_content = "\n\n".join(knowledge_base["document_content"])
|
106 |
+
url_content = "\n\n".join(knowledge_base["url_content"])
|
107 |
+
|
108 |
+
internal_prompt = """
|
109 |
+
Instructions for the model:
|
110 |
+
- Follow the principles of news writing: always try to answer the 5 Ws of a news story in the first paragraph (Who?, What?, When?, Where?, Why?).
|
111 |
+
- Ensure that at least 80% of the quotes are direct and in quotation marks.
|
112 |
+
- The remaining 20% can be indirect quotes.
|
113 |
+
- Do not invent new information.
|
114 |
+
- Be rigorous with the provided facts.
|
115 |
+
- When processing uploaded documents, extract and highlight important quotes and verbatim testimonies from sources.
|
116 |
+
- When processing uploaded documents, extract and highlight key figures.
|
117 |
"""
|
118 |
|
119 |
prompt = f"""
|
120 |
+
{internal_prompt}
|
121 |
+
Write a news article with the following information, including a title, a 15-word hook (additional information that complements the title), and the body content with a size of {size} words. The tone should be {tone}.
|
122 |
+
Instructions: {knowledge_base["instructions"]}
|
123 |
+
Facts: {knowledge_base["facts"]}
|
124 |
+
Additional content from documents: {document_content}
|
125 |
+
Additional content from URLs: {url_content}
|
126 |
+
Use the following transcriptions as direct and indirect quotes (without changing or inventing content):
|
127 |
+
{transcriptions_text}
|
128 |
"""
|
129 |
|
130 |
try:
|
131 |
+
response = openai.ChatCompletion.create(
|
132 |
model="gpt-3.5-turbo",
|
133 |
messages=[{"role": "user", "content": prompt}],
|
134 |
temperature=0.1
|
135 |
)
|
136 |
+
news_article = response['choices'][0]['message']['content']
|
137 |
+
return news_article, raw_transcriptions
|
138 |
except Exception as e:
|
139 |
+
return f"Error generating the news article: {str(e)}", ""
|
140 |
|
141 |
with gr.Blocks() as demo:
|
142 |
+
gr.Markdown("## All-in-One News Generator")
|
143 |
with gr.Row():
|
144 |
with gr.Column(scale=2):
|
145 |
+
instructions = gr.Textbox(label="Instructions for the news article", lines=2)
|
146 |
+
facts = gr.Textbox(label="Describe the facts of the news", lines=4)
|
147 |
+
size = gr.Number(label="Size of the news body (in words)", value=100)
|
148 |
+
tone = gr.Dropdown(label="Tone of the news", choices=["serious", "neutral", "lighthearted"], value="neutral")
|
149 |
+
urls = gr.Textbox(label="URLs (separated by space)", lines=2)
|
150 |
with gr.Column(scale=3):
|
151 |
+
inputs_list = [instructions, facts, size, tone, urls]
|
152 |
with gr.Tabs():
|
153 |
for i in range(1, 6):
|
154 |
with gr.TabItem(f"Audio {i}"):
|
155 |
audio = gr.Audio(type="filepath", label=f"Audio {i}")
|
156 |
+
name = gr.Textbox(label="Name", scale=1)
|
157 |
+
position = gr.Textbox(label="Position", scale=1)
|
158 |
+
inputs_list.extend([audio, name, position])
|
159 |
for i in range(1, 6):
|
160 |
+
with gr.TabItem(f"Document {i}"):
|
161 |
+
document = gr.File(label=f"Document {i}", type="filepath", file_count="single")
|
162 |
+
inputs_list.append(document)
|
163 |
|
164 |
+
gr.Markdown("---") # Visual separator
|
165 |
|
166 |
with gr.Row():
|
167 |
+
transcriptions_output = gr.Textbox(label="Transcriptions", lines=10)
|
168 |
|
169 |
+
gr.Markdown("---") # Visual separator
|
170 |
|
171 |
with gr.Row():
|
172 |
+
generate = gr.Button("Generate draft")
|
173 |
with gr.Row():
|
174 |
+
news_output = gr.Textbox(label="Generated draft", lines=20)
|
175 |
|
176 |
+
generate.click(fn=generate_news, inputs=inputs_list, outputs=[news_output, transcriptions_output])
|
177 |
|
178 |
+
demo.launch(share=True)
|