Spaces:
Running
Running
import gradio as gr | |
from gradio_client import Client | |
import json | |
import logging | |
import ast | |
import openai | |
import os | |
import random | |
import re | |
logging.basicConfig(filename='youtube_script_extractor.log', level=logging.DEBUG, | |
format='%(asctime)s - %(levelname)s - %(message)s') | |
openai.api_key = os.getenv("OPENAI_API_KEY") | |
def parse_api_response(response): | |
try: | |
if isinstance(response, str): | |
response = ast.literal_eval(response) | |
if isinstance(response, list) and len(response) > 0: | |
response = response[0] | |
if not isinstance(response, dict): | |
raise ValueError(f"μμμΉ λͺ»ν μλ΅ νμμ λλ€. λ°μ λ°μ΄ν° νμ : {type(response)}") | |
return response | |
except Exception as e: | |
raise ValueError(f"API μλ΅ νμ± μ€ν¨: {str(e)}") | |
def split_sentences(text): | |
sentences = re.split(r"(λλ€|μμ|ꡬλ|ν΄μ|κ΅°μ|κ² μ΄μ|μμ€|ν΄λΌ|μμ|μμ|λ°μ|λμ|μΈμ|μ΄μ|κ²μ|ꡬμ|κ³ μ|λμ|νμ£ )(?![\w])", text) | |
combined_sentences = [] | |
current_sentence = "" | |
for i in range(0, len(sentences), 2): | |
if i + 1 < len(sentences): | |
sentence = sentences[i] + sentences[i + 1] | |
else: | |
sentence = sentences[i] | |
if len(current_sentence) + len(sentence) > 100: | |
combined_sentences.append(current_sentence.strip()) | |
current_sentence = sentence.strip() | |
else: | |
current_sentence += sentence | |
if sentence.endswith(('.', '?', '!')): | |
combined_sentences.append(current_sentence.strip()) | |
current_sentence = "" | |
if current_sentence: | |
combined_sentences.append(current_sentence.strip()) | |
return combined_sentences | |
def get_youtube_script(url): | |
logging.info(f"μ€ν¬λ¦½νΈ μΆμΆ μμ: URL = {url}") | |
client = Client("whispersound/YT_Ts_R") | |
try: | |
logging.debug("API νΈμΆ μμ") | |
result = client.predict(youtube_url=url, api_name="/predict") | |
logging.debug("API νΈμΆ μλ£") | |
parsed_result = parse_api_response(result) | |
title = parsed_result["data"][0]["title"] | |
transcription_text = parsed_result["data"][0]["transcriptionAsText"] | |
sections = parsed_result["data"][0]["sections"] | |
logging.info("μ€ν¬λ¦½νΈ μΆμΆ μλ£") | |
return title, transcription_text, sections | |
except Exception as e: | |
error_msg = f"μ€ν¬λ¦½νΈ μΆμΆ μ€ μ€λ₯ λ°μ: {str(e)}" | |
logging.exception(error_msg) | |
return "", "", [] | |
def call_api(prompt, max_tokens, temperature, top_p): | |
try: | |
response = openai.ChatCompletion.create( | |
model="gpt-4o-mini", | |
messages=[{"role": "user", "content": prompt}], | |
max_tokens=max_tokens, | |
temperature=temperature, | |
top_p=top_p | |
) | |
return response['choices'][0]['message']['content'] | |
except Exception as e: | |
logging.exception("LLM API νΈμΆ μ€ μ€λ₯ λ°μ") | |
return "μμ½μ μμ±νλ λμ μ€λ₯κ° λ°μνμ΅λλ€. λμ€μ λ€μ μλν΄ μ£ΌμΈμ." | |
def summarize_section(section_text): | |
prompt = f""" | |
λ€μ μ νλΈ λλ³Έ μΉμ μ ν΅μ¬ λ΄μ©μ κ°κ²°νκ² μμ½νμΈμ: | |
1. νκΈλ‘ μμ±νμΈμ. | |
2. μ£Όμ λ Όμ κ³Ό μ€μν μΈλΆμ¬νμ ν¬ν¨νμΈμ. | |
3. μμ½μ 2-3λ¬Έμ₯μΌλ‘ μ ννμΈμ. | |
μΉμ λ΄μ©: | |
{section_text} | |
""" | |
return call_api(prompt, max_tokens=150, temperature=0.3, top_p=0.9) | |
def format_time(seconds): | |
minutes, seconds = divmod(seconds, 60) | |
hours, minutes = divmod(minutes, 60) | |
return f"{int(hours):02d}:{int(minutes):02d}:{int(seconds):02d}" | |
def generate_timeline_summary(sections): | |
timeline_summary = "" | |
for i, section in enumerate(sections, 1): | |
start_time = format_time(section['start_time']) | |
summary = summarize_section(section['text']) | |
timeline_summary += f"{start_time} {i}. {summary}\n\n" | |
return timeline_summary | |
def summarize_text(text): | |
prompt = f""" | |
1. λ€μ μ£Όμ΄μ§λ μ νλΈ λλ³Έμ ν΅μ¬ μ£Όμ μ λͺ¨λ μ£Όμ λ΄μ©μ μμΈνκ² μμ½νλΌ | |
2. λ°λμ νκΈλ‘ μμ±νλΌ | |
3. μμ½λ¬Έλ§μΌλ‘λ μμμ μ§μ μμ²ν κ²κ³Ό λμΌν μμ€μΌλ‘ λ΄μ©μ μ΄ν΄ν μ μλλ‘ μμΈν μμ± | |
4. κΈμ λ무 μμΆνκ±°λ ν¨μΆνμ§ λ§κ³ , μ€μν λ΄μ©κ³Ό μΈλΆμ¬νμ λͺ¨λ ν¬ν¨ | |
5. λ°λμ λλ³Έμ νλ¦κ³Ό λ Όλ¦¬ ꡬ쑰λ₯Ό μ μ§ | |
6. λ°λμ μκ° μμλ μ¬κ±΄μ μ κ° κ³Όμ μ λͺ ννκ² λ°μ | |
7. λ±μ₯μΈλ¬Ό, μ₯μ, μ¬κ±΄ λ± μ€μν μμλ₯Ό μ ννκ² μμ± | |
8. λλ³Έμμ μ λ¬νλ κ°μ μ΄λ λΆμκΈ°λ ν¬ν¨ | |
9. λ°λμ κΈ°μ μ μ©μ΄λ μ λ¬Έ μ©μ΄κ° μμ κ²½μ°, μ΄λ₯Ό μ ννκ² μ¬μ© | |
10. λλ³Έμ λͺ©μ μ΄λ μλλ₯Ό νμ νκ³ , μ΄λ₯Ό μμ½μ λ°λμ λ°μ | |
11. μ 체κΈμ λ³΄κ³ | |
--- | |
μ΄ ν둬ννΈκ° λμμ΄ λμκΈΈ λ°λλλ€. | |
\n\n | |
{text}""" | |
try: | |
return call_api(prompt, max_tokens=10000, temperature=0.3, top_p=0.9) | |
except Exception as e: | |
logging.exception("μμ½ μμ± μ€ μ€λ₯ λ°μ") | |
return "μμ½μ μμ±νλ λμ μ€λ₯κ° λ°μνμ΅λλ€. λμ€μ λ€μ μλν΄ μ£ΌμΈμ." | |
with gr.Blocks() as demo: | |
gr.Markdown("## YouTube μ€ν¬λ¦½νΈ μΆμΆ λ° μμ½ λꡬ") | |
youtube_url_input = gr.Textbox(label="YouTube URL μ λ ₯") | |
analyze_button = gr.Button("λΆμνκΈ°") | |
script_output = gr.HTML(label="μ€ν¬λ¦½νΈ") | |
timeline_output = gr.HTML(label="νμλΌμΈ μμ½") | |
summary_output = gr.HTML(label="μ 체 μμ½") | |
cached_data = gr.State({"url": "", "title": "", "script": "", "sections": []}) | |
def extract_and_cache(url, cache): | |
if url == cache["url"]: | |
return cache["title"], cache["script"], cache["sections"], cache | |
title, script, sections = get_youtube_script(url) | |
new_cache = {"url": url, "title": title, "script": script, "sections": sections} | |
return title, script, sections, new_cache | |
def display_script(title, script): | |
if not script: | |
return "<p>μ€ν¬λ¦½νΈλ₯Ό μΆμΆνμ§ λͺ»νμ΅λλ€. URLμ νμΈνκ³ λ€μ μλν΄ μ£ΌμΈμ.</p>" | |
formatted_script = "\n".join(split_sentences(script)) | |
script_html = f"""<h2 style='font-size:24px;'>{title}</h2> | |
<details> | |
<summary><h3>μλ¬Έ μ€ν¬λ¦½νΈ (ν΄λ¦νμ¬ νΌμΉκΈ°)</h3></summary> | |
<div style="white-space: pre-wrap;">{formatted_script}</div> | |
</details>""" | |
return script_html | |
def display_timeline(sections): | |
if not sections: | |
return "<p>νμλΌμΈμ μμ±νμ§ λͺ»νμ΅λλ€. μ€ν¬λ¦½νΈ μΆμΆμ μ€ν¨νμ μ μμ΅λλ€.</p>" | |
timeline_summary = generate_timeline_summary(sections) | |
timeline_html = f""" | |
<h3>νμλΌμΈ μμ½:</h3> | |
<div style="white-space: pre-wrap; max-height: 400px; overflow-y: auto; border: 1px solid #ccc; padding: 10px;"> | |
{timeline_summary} | |
</div> | |
""" | |
return timeline_html | |
def generate_summary(script): | |
if not script: | |
return "<p>μ 체 μμ½μ μμ±νμ§ λͺ»νμ΅λλ€. μ€ν¬λ¦½νΈ μΆμΆμ μ€ν¨νμ μ μμ΅λλ€.</p>" | |
summary = summarize_text(script) | |
summary_html = f""" | |
<h3>μ 체 μμ½:</h3> | |
<div style="white-space: pre-wrap; max-height: 400px; overflow-y: auto; border: 1px solid #ccc; padding: 10px;"> | |
{summary} | |
</div> | |
""" | |
return summary_html | |
def analyze(url, cache): | |
title, script, sections, new_cache = extract_and_cache(url, cache) | |
script_html = display_script(title, script) | |
timeline_html = display_timeline(sections) | |
summary_html = generate_summary(script) | |
return script_html, timeline_html, summary_html, new_cache | |
analyze_button.click( | |
analyze, | |
inputs=[youtube_url_input, cached_data], | |
outputs=[script_output, timeline_output, summary_output, cached_data] | |
) | |
demo.launch(share=True) |