Spaces:
Sleeping
Sleeping
final
Browse files- app.py +407 -0
- requirements.txt +12 -0
app.py
ADDED
@@ -0,0 +1,407 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import zipfile
|
3 |
+
import requests
|
4 |
+
import gradio as gr
|
5 |
+
import whisper
|
6 |
+
import subprocess
|
7 |
+
import uuid
|
8 |
+
import torch
|
9 |
+
import re
|
10 |
+
import matplotlib.pyplot as plt
|
11 |
+
import language_tool_python
|
12 |
+
import difflib
|
13 |
+
from transformers import (
|
14 |
+
AutoTokenizer,
|
15 |
+
AutoModelForSeq2SeqLM,
|
16 |
+
pipeline as hf_pipeline,
|
17 |
+
)
|
18 |
+
|
19 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
20 |
+
# Optional evaluation libraries
|
21 |
+
try:
|
22 |
+
from rouge_score import rouge_scorer
|
23 |
+
except ImportError:
|
24 |
+
rouge_scorer = None
|
25 |
+
print("[Warning] rouge_score ν¨ν€μ§κ° μμ΅λλ€. pip install rouge-score")
|
26 |
+
|
27 |
+
try:
|
28 |
+
from bert_score import score as bert_score_func
|
29 |
+
except ImportError:
|
30 |
+
bert_score_func = None
|
31 |
+
print("[Warning] bert-score ν¨ν€μ§κ° μμ΅λλ€. pip install bert-score")
|
32 |
+
|
33 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
34 |
+
# νκΈ λ§μΆ€λ² κ²μ¬(pyβhanspell)
|
35 |
+
try:
|
36 |
+
from hanspell import spell_checker
|
37 |
+
except ImportError:
|
38 |
+
spell_checker = None
|
39 |
+
|
40 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
41 |
+
# LanguageTool λ£° κΈ°λ° κ΅μ (μμ΄ μ μ©)
|
42 |
+
try:
|
43 |
+
lt_tool = language_tool_python.LanguageTool('en-US')
|
44 |
+
except Exception as e:
|
45 |
+
lt_tool = None
|
46 |
+
print(f"[Warning] LanguageTool μ΄κΈ°ν μ€ν¨: {e}")
|
47 |
+
|
48 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
49 |
+
# FFmpeg
|
50 |
+
yt_dlp_path = "C:/Windows/System32/yt-dlp.exe"
|
51 |
+
ffmpeg_path = "C:/ProgramData/chocolatey/bin"
|
52 |
+
def download_ffmpeg(dest_bin):
|
53 |
+
if os.path.isdir(dest_bin) and os.path.isfile(os.path.join(dest_bin, "ffmpeg.exe")):
|
54 |
+
return dest_bin
|
55 |
+
url = "https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-essentials.zip"
|
56 |
+
zip_path = os.path.join(os.getcwd(), "ffmpeg.zip")
|
57 |
+
extract_root = os.path.dirname(dest_bin)
|
58 |
+
os.makedirs(extract_root, exist_ok=True)
|
59 |
+
resp = requests.get(url, stream=True); resp.raise_for_status()
|
60 |
+
with open(zip_path, "wb") as f:
|
61 |
+
for chunk in resp.iter_content(8192): f.write(chunk)
|
62 |
+
with zipfile.ZipFile(zip_path, "r") as zf: zf.extractall(extract_root)
|
63 |
+
os.remove(zip_path)
|
64 |
+
for root, _, files in os.walk(extract_root):
|
65 |
+
if "ffmpeg.exe" in files:
|
66 |
+
os.makedirs(dest_bin, exist_ok=True)
|
67 |
+
for fn in ("ffmpeg.exe","ffprobe.exe","ffplay.exe"):
|
68 |
+
src, dst = os.path.join(root,fn), os.path.join(dest_bin,fn)
|
69 |
+
if os.path.isfile(src): os.replace(src, dst)
|
70 |
+
return dest_bin
|
71 |
+
raise RuntimeError("FFmpeg μ€μΉ μ€ν¨")
|
72 |
+
|
73 |
+
download_ffmpeg(ffmpeg_path)
|
74 |
+
os.environ["PATH"] = ffmpeg_path + os.pathsep + os.environ.get("PATH","")
|
75 |
+
|
76 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
77 |
+
# Whisper
|
78 |
+
asr_model = whisper.load_model("medium")
|
79 |
+
|
80 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
81 |
+
# μμ½ λͺ¨λΈ(λͺ¨λΈ/ν ν¬λμ΄μ μ§μ μ¬μ©, pipeline X)
|
82 |
+
SUMMARY_MODELS = {
|
83 |
+
"mT5_multilingual_XLSum": "csebuetnlp/mT5_multilingual_XLSum",
|
84 |
+
"Pegasus XSum": "google/pegasus-xsum",
|
85 |
+
"BART-large CNN": "facebook/bart-large-cnn",
|
86 |
+
"DistilBART CNN": "sshleifer/distilbart-cnn-12-6"
|
87 |
+
}
|
88 |
+
tokenizers, models = {}, {}
|
89 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
90 |
+
|
91 |
+
def load_summarizer(label: str):
|
92 |
+
if label in models:
|
93 |
+
return
|
94 |
+
repo = SUMMARY_MODELS[label]
|
95 |
+
tok = AutoTokenizer.from_pretrained(repo, use_fast=False)
|
96 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(repo).to(device)
|
97 |
+
model.eval()
|
98 |
+
tokenizers[label] = tok
|
99 |
+
models[label] = model
|
100 |
+
|
101 |
+
if rouge_scorer:
|
102 |
+
scorer = rouge_scorer.RougeScorer(["rouge1","rouge2","rougeL"], use_stemmer=True)
|
103 |
+
|
104 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
105 |
+
# λ¬Έλ² κ΅μ
|
106 |
+
GRAMMAR_MODELS = {
|
107 |
+
"LanguageTool-en": None,
|
108 |
+
"py-hanspell": None,
|
109 |
+
"GEC-νκ΅μ΄": "Soyoung97/gec_kr"
|
110 |
+
}
|
111 |
+
grammar_pipes = {}
|
112 |
+
|
113 |
+
def load_grammar_pipe(name: str):
|
114 |
+
repo = GRAMMAR_MODELS[name]
|
115 |
+
grammar_pipes[name] = hf_pipeline(
|
116 |
+
"text2text-generation",
|
117 |
+
model=repo,
|
118 |
+
tokenizer=AutoTokenizer.from_pretrained(repo),
|
119 |
+
device=0 if torch.cuda.is_available() else -1
|
120 |
+
)
|
121 |
+
|
122 |
+
def correct_spelling(text, max_chunk=500):
|
123 |
+
if not spell_checker: return text
|
124 |
+
parts, curr = re.split(r'([.?!]\s*)', text), ""
|
125 |
+
segs, out = [], []
|
126 |
+
for p in parts:
|
127 |
+
if len(curr)+len(p) <= max_chunk: curr += p
|
128 |
+
else: segs.append(curr); curr = p
|
129 |
+
if curr: segs.append(curr)
|
130 |
+
for s in segs:
|
131 |
+
try: out.append(spell_checker.check(s).checked)
|
132 |
+
except: out.append(s)
|
133 |
+
return " ".join(o.strip() for o in out)
|
134 |
+
|
135 |
+
def correct_text(text, method="GEC-νκ΅μ΄"):
|
136 |
+
if method=="py-hanspell":
|
137 |
+
return correct_spelling(text)
|
138 |
+
if method=="LanguageTool-en" and lt_tool:
|
139 |
+
matches = lt_tool.check(text)
|
140 |
+
return language_tool_python.utils.correct(text, matches)
|
141 |
+
if method=="GEC-νκ΅μ΄":
|
142 |
+
if method not in grammar_pipes:
|
143 |
+
load_grammar_pipe(method)
|
144 |
+
pipe = grammar_pipes[method]
|
145 |
+
sents = re.split(r'(?<=[.?!])\s+', text)
|
146 |
+
corrected=[]
|
147 |
+
for sent in sents:
|
148 |
+
gen = pipe(sent, max_length=256, min_length=1, do_sample=False)[0]["generated_text"]
|
149 |
+
corrected.append(gen.strip())
|
150 |
+
return " ".join(corrected)
|
151 |
+
return text
|
152 |
+
|
153 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
154 |
+
# κ΅μ λ₯ + Diff
|
155 |
+
def calculate_correction_rate(original, corrected):
|
156 |
+
orig_tokens = original.split()
|
157 |
+
corr_tokens = corrected.split()
|
158 |
+
sm = difflib.SequenceMatcher(None, orig_tokens, corr_tokens)
|
159 |
+
diff_count = sum((i2 - i1) for tag, i1, i2, j1, j2 in sm.get_opcodes() if tag != 'equal')
|
160 |
+
total = max(len(orig_tokens), 1)
|
161 |
+
return round(100 * diff_count / total, 2)
|
162 |
+
|
163 |
+
def highlight_diff(original: str, corrected: str) -> str:
|
164 |
+
diff = difflib.ndiff(original.split(), corrected.split())
|
165 |
+
html_parts = []
|
166 |
+
for token in diff:
|
167 |
+
if token.startswith("+ "):
|
168 |
+
html_parts.append(f"<span style='color:red;'>{token[2:]}</span>")
|
169 |
+
elif token.startswith("- "):
|
170 |
+
continue
|
171 |
+
else:
|
172 |
+
html_parts.append(token[2:])
|
173 |
+
return " ".join(html_parts)
|
174 |
+
|
175 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
176 |
+
# YouTube
|
177 |
+
def download_audio(url):
|
178 |
+
fname = f"yt_{uuid.uuid4().hex[:8]}.mp3"
|
179 |
+
cmd = [yt_dlp_path,"-f","bestaudio","--extract-audio","--audio-format","mp3","-o",fname,url]
|
180 |
+
res = subprocess.run(cmd, capture_output=True, text=True)
|
181 |
+
if res.returncode!=0: raise RuntimeError(res.stderr)
|
182 |
+
return fname
|
183 |
+
|
184 |
+
def get_transcript(url, state):
|
185 |
+
if state and state.get("url")==url:
|
186 |
+
return state["orig"], state
|
187 |
+
audio = download_audio(url)
|
188 |
+
res = asr_model.transcribe(audio)
|
189 |
+
orig = res.get("text","")
|
190 |
+
os.remove(audio)
|
191 |
+
return orig, {"url":url, "orig":orig}
|
192 |
+
|
193 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
194 |
+
# μμ ν μ²ν¬ μμ½ (model.generate μ§μ νΈμΆ)
|
195 |
+
def summarize_long_text(text: str, label: str, chunk_size: int = 512) -> str:
|
196 |
+
load_summarizer(label)
|
197 |
+
tok = tokenizers[label]
|
198 |
+
model= models[label]
|
199 |
+
|
200 |
+
enc = tok(text, return_tensors="pt", truncation=False)
|
201 |
+
ids = enc.input_ids[0]
|
202 |
+
summaries = []
|
203 |
+
|
204 |
+
max_ctx = getattr(model.config, "max_position_embeddings", 1024) - 4
|
205 |
+
chunk_size = min(chunk_size, max_ctx)
|
206 |
+
|
207 |
+
for i in range(0, len(ids), chunk_size):
|
208 |
+
chunk_ids = ids[i:i+chunk_size].unsqueeze(0).to(device)
|
209 |
+
out_ids = model.generate(
|
210 |
+
chunk_ids,
|
211 |
+
max_new_tokens=128,
|
212 |
+
num_beams=4,
|
213 |
+
do_sample=False
|
214 |
+
)
|
215 |
+
summ = tok.decode(out_ids[0], skip_special_tokens=True)
|
216 |
+
summaries.append(summ)
|
217 |
+
|
218 |
+
combined = " ".join(summaries)
|
219 |
+
enc2 = tok(combined, return_tensors="pt", truncation=True, max_length=max_ctx).to(device)
|
220 |
+
out_ids = model.generate(
|
221 |
+
**enc2,
|
222 |
+
max_new_tokens=128,
|
223 |
+
num_beams=4,
|
224 |
+
do_sample=False
|
225 |
+
)
|
226 |
+
final = tok.decode(out_ids[0], skip_special_tokens=True)
|
227 |
+
return final
|
228 |
+
|
229 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
230 |
+
def summarize_single(url, label, grammar_method, transcript_state):
|
231 |
+
orig, new_state = get_transcript(url, transcript_state)
|
232 |
+
corr = correct_text(orig, method=grammar_method)
|
233 |
+
corr_rate = calculate_correction_rate(orig, corr)
|
234 |
+
corr_html = f"<div><b>κ΅μ λ₯ :</b> {corr_rate}%</div><hr/>{highlight_diff(orig, corr)}"
|
235 |
+
|
236 |
+
summary = summarize_long_text(corr, label) if len(corr) > 100 else "β οΈ μμ½ λΆκ°"
|
237 |
+
|
238 |
+
rouge_vals=[0,0,0]
|
239 |
+
if rouge_scorer and summary.strip():
|
240 |
+
sc = scorer.score(orig, summary)
|
241 |
+
rouge_vals=[sc["rouge1"].fmeasure, sc["rouge2"].fmeasure, sc["rougeL"].fmeasure]
|
242 |
+
|
243 |
+
bert_f1=0
|
244 |
+
if bert_score_func and summary.strip():
|
245 |
+
try:
|
246 |
+
_,_,F = bert_score_func([summary],[orig],lang="ko")
|
247 |
+
except Exception:
|
248 |
+
_,_,F = bert_score_func([summary],[orig],lang="en")
|
249 |
+
bert_f1=float(F.mean())
|
250 |
+
|
251 |
+
fig,ax=plt.subplots()
|
252 |
+
ax.bar(["R1","R2","RL","BERT-F1"], rouge_vals+[bert_f1])
|
253 |
+
ax.set_ylim(0,1); ax.set_ylabel("Score"); ax.set_title("Summary Fidelity")
|
254 |
+
plt.tight_layout()
|
255 |
+
|
256 |
+
return orig, corr_html, summary, fig, new_state
|
257 |
+
|
258 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
259 |
+
def summarize_all(url, grammar_method, transcript_state):
|
260 |
+
orig, new_state = get_transcript(url, transcript_state)
|
261 |
+
corr = correct_text(orig, method=grammar_method)
|
262 |
+
corr_rate = calculate_correction_rate(orig, corr)
|
263 |
+
corr_html = f"<div><b>κ΅μ λ₯ :</b> {corr_rate}%</div><hr/>{highlight_diff(orig, corr)}"
|
264 |
+
|
265 |
+
figs, interps, rv_list, bf_list = [], [], [], []
|
266 |
+
summaries_plain = []
|
267 |
+
labels = list(SUMMARY_MODELS.keys())
|
268 |
+
|
269 |
+
for label in labels:
|
270 |
+
summ = summarize_long_text(corr, label)
|
271 |
+
summaries_plain.append(summ)
|
272 |
+
|
273 |
+
rv=[0,0,0]; bf=0
|
274 |
+
if rouge_scorer:
|
275 |
+
sc = scorer.score(orig, summ)
|
276 |
+
rv=[sc["rouge1"].fmeasure, sc["rouge2"].fmeasure, sc["rougeL"].fmeasure]
|
277 |
+
if bert_score_func:
|
278 |
+
try:
|
279 |
+
_,_,F = bert_score_func([summ],[orig],lang="ko")
|
280 |
+
except Exception:
|
281 |
+
_,_,F = bert_score_func([summ],[orig],lang="en")
|
282 |
+
bf=float(F.mean())
|
283 |
+
rv_list.append(rv); bf_list.append(bf)
|
284 |
+
|
285 |
+
fig,ax=plt.subplots()
|
286 |
+
ax.bar(["R1","R2","RL","BERT-F1"], rv+[bf])
|
287 |
+
ax.set_ylim(0,1); ax.set_title(label)
|
288 |
+
plt.tight_layout(); figs.append(fig)
|
289 |
+
|
290 |
+
note="μ 보 μμ€ λ§μ"
|
291 |
+
if bf>0.8: note="ν΅μ¬ μ 보 μ λ°μ"
|
292 |
+
elif bf>0.5: note="μ£Όμ λ΄μ© ν¬ν¨"
|
293 |
+
interps.append(f"{label}: {note} (F1={bf:.2f})")
|
294 |
+
|
295 |
+
html = "<h3>λͺ¨λΈλ³ μμ½ & Fidelity Metrics</h3>"
|
296 |
+
html+= f"<p><b>κ΅μ λ₯ :</b> {corr_rate}%</p>"
|
297 |
+
html+= "<table border='1' style='border-collapse:collapse; width:100%; table-layout:fixed;'>"
|
298 |
+
html+= "<tr><th style='width:12%'>λͺ¨λΈ</th><th style='width:58%'>μμ½λ¬Έ</th><th style='width:5%'>R1</th><th style='width:5%'>R2</th><th style='width:5%'>RL</th><th style='width:7%'>BERT-F1</th><th style='width:8%'>ν΄μ</th></tr>"
|
299 |
+
|
300 |
+
for i,label in enumerate(labels):
|
301 |
+
r1,r2,rl = rv_list[i]
|
302 |
+
bf = bf_list[i]
|
303 |
+
note = "μ 보 μμ€ λ§μ"
|
304 |
+
if bf>0.8: note="ν΅μ¬ μ 보 μ λ°μ"
|
305 |
+
elif bf>0.5: note="μ£Όμ λ΄μ© ν¬ν¨"
|
306 |
+
|
307 |
+
summ_html = summaries_plain[i].replace("<", "<")
|
308 |
+
html+= (
|
309 |
+
f"<tr>"
|
310 |
+
f"<td>{label}</td>"
|
311 |
+
f"<td style='white-space:pre-wrap; word-break:break-word'>{summ_html}</td>"
|
312 |
+
f"<td>{r1:.2f}</td><td>{r2:.2f}</td><td>{rl:.2f}</td>"
|
313 |
+
f"<td>{bf:.2f}</td><td>{note}</td>"
|
314 |
+
f"</tr>"
|
315 |
+
)
|
316 |
+
html+="</table>"
|
317 |
+
|
318 |
+
return [orig, corr_html] + figs + interps + [html, new_state]
|
319 |
+
|
320 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
321 |
+
def save_summary(url, label):
|
322 |
+
orig, _ = get_transcript(url, None)
|
323 |
+
corr = correct_text(orig, "GEC-νκ΅μ΄")
|
324 |
+
summary = summarize_long_text(corr, label)
|
325 |
+
path = os.path.join(os.getcwd(), f"summary_{label}.txt")
|
326 |
+
with open(path, "w", encoding="utf-8") as f:
|
327 |
+
f.write(summary)
|
328 |
+
return path
|
329 |
+
|
330 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
331 |
+
# CSS (κ΅μ μλ§μ λ°μ€μ²λΌ 보μ΄κ²)
|
332 |
+
CUSTOM_CSS = """
|
333 |
+
#corr_box, #corr_box_all {
|
334 |
+
border: 1px solid #ccc;
|
335 |
+
padding: 10px;
|
336 |
+
border-radius: 6px;
|
337 |
+
background-color: #fafafa;
|
338 |
+
max-height: 300px;
|
339 |
+
overflow-y: auto;
|
340 |
+
white-space: pre-wrap;
|
341 |
+
}
|
342 |
+
"""
|
343 |
+
|
344 |
+
# Gradio
|
345 |
+
with gr.Blocks(css=CUSTOM_CSS) as demo:
|
346 |
+
gr.Markdown("## π¬ YouTube μμ½ μλΉμ€ (κ΅μ + κ΅μ λ₯ + Diff κ°μ‘°, μμ μ²ν¬μμ½)")
|
347 |
+
|
348 |
+
with gr.Tabs():
|
349 |
+
with gr.TabItem("λ¨μΌ λͺ¨λΈ μμ½"):
|
350 |
+
url_input = gr.Textbox(label="YouTube URL")
|
351 |
+
model_sel = gr.Dropdown(list(SUMMARY_MODELS.keys()), label="μμ½ λͺ¨λΈ")
|
352 |
+
grammar_sel = gr.Dropdown(list(GRAMMAR_MODELS.keys()), label="κ΅μ λͺ¨λΈ", value="GEC-νκ΅μ΄")
|
353 |
+
transcript_state = gr.State(None)
|
354 |
+
btn_single = gr.Button("μμ½ μ€ν")
|
355 |
+
|
356 |
+
orig_tb = gr.Textbox(label="μλ¬Έ μλ§", lines=10)
|
357 |
+
corr_tb = gr.HTML(label="κ΅μ μλ§ (λ³κ²½μ κ°μ‘°)", elem_id="corr_box")
|
358 |
+
sum_tb = gr.Textbox(label="μμ½ κ²°κ³Ό", lines=8)
|
359 |
+
fidelity_plot = gr.Plot(label="Fidelity Metrics")
|
360 |
+
save_btn = gr.Button("μμ½ μ μ₯")
|
361 |
+
download_single = gr.File(label="λ€μ΄λ‘λ νμΌ")
|
362 |
+
|
363 |
+
btn_single.click(
|
364 |
+
fn=summarize_single,
|
365 |
+
inputs=[url_input, model_sel, grammar_sel, transcript_state],
|
366 |
+
outputs=[orig_tb, corr_tb, sum_tb, fidelity_plot, transcript_state]
|
367 |
+
)
|
368 |
+
save_btn.click(
|
369 |
+
fn=save_summary,
|
370 |
+
inputs=[url_input, model_sel],
|
371 |
+
outputs=[download_single]
|
372 |
+
)
|
373 |
+
|
374 |
+
with gr.TabItem("μ 체 λͺ¨λΈ λΉκ΅"):
|
375 |
+
url_all = gr.Textbox(label="YouTube URL")
|
376 |
+
grammar_sel_all = gr.Dropdown(list(GRAMMAR_MODELS.keys()), label="κ΅μ λͺ¨λΈ", value="GEC-νκ΅μ΄")
|
377 |
+
transcript_state_all = gr.State(None)
|
378 |
+
btn_all = gr.Button("λͺ¨λ μ€ν")
|
379 |
+
|
380 |
+
orig_all = gr.Textbox(label="μλ¬Έ μλ§", lines=10)
|
381 |
+
corr_all = gr.HTML(label="κ΅μ μλ§ (λ³κ²½μ κ°μ‘°)", elem_id="corr_box_all")
|
382 |
+
|
383 |
+
plot_components, interp_components = [], []
|
384 |
+
for label in SUMMARY_MODELS:
|
385 |
+
plot_components.append(gr.Plot(label=f"{label} Metrics"))
|
386 |
+
interp_components.append(gr.HTML(label=f"{label} ν΄μ"))
|
387 |
+
|
388 |
+
agg_table = gr.HTML(label="λͺ¨λΈλ³ μμ½ & Fidelity Metrics")
|
389 |
+
save_all_sel = gr.Radio(list(SUMMARY_MODELS.keys()), label="μ μ₯ λͺ¨λΈ μ§μ ")
|
390 |
+
save_all_btn = gr.Button("μ ν μμ½ μ μ₯")
|
391 |
+
download_all = gr.File(label="λ€μ΄λ‘λ νμΌ")
|
392 |
+
|
393 |
+
btn_all.click(
|
394 |
+
fn=summarize_all,
|
395 |
+
inputs=[url_all, grammar_sel_all, transcript_state_all],
|
396 |
+
outputs=[orig_all, corr_all] + plot_components + interp_components + [agg_table, transcript_state_all]
|
397 |
+
)
|
398 |
+
save_all_btn.click(
|
399 |
+
fn=save_summary,
|
400 |
+
inputs=[url_all, save_all_sel],
|
401 |
+
outputs=[download_all]
|
402 |
+
)
|
403 |
+
|
404 |
+
if __name__ == '__main__':
|
405 |
+
# μλ ν¬νΈ ν λΉ
|
406 |
+
demo.launch(server_name="127.0.0.1")
|
407 |
+
# νΉμ μμ μλ: demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
transformers
|
3 |
+
sentencepiece
|
4 |
+
gradio
|
5 |
+
git+https://github.com/openai/whisper.git
|
6 |
+
matplotlib
|
7 |
+
requests
|
8 |
+
uuid
|
9 |
+
language-tool-python
|
10 |
+
rouge-score
|
11 |
+
bert-score
|
12 |
+
yt-dlp
|