Update app.py
Browse files
app.py
CHANGED
|
@@ -4,12 +4,14 @@ import shutil
|
|
| 4 |
from pydub import AudioSegment
|
| 5 |
from openai import OpenAI
|
| 6 |
import gradio as gr
|
| 7 |
-
from fastapi import
|
|
|
|
|
|
|
| 8 |
|
| 9 |
# ======================================================
|
| 10 |
# 🔐 設定區
|
| 11 |
# ======================================================
|
| 12 |
-
PASSWORD = os.getenv("APP_PASSWORD", "
|
| 13 |
MAX_SIZE = 25 * 1024 * 1024
|
| 14 |
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
| 15 |
|
|
@@ -25,77 +27,82 @@ def split_audio_if_needed(path):
|
|
| 25 |
if size <= MAX_SIZE:
|
| 26 |
return [path]
|
| 27 |
audio = AudioSegment.from_file(path)
|
| 28 |
-
|
| 29 |
-
chunk_ms = len(audio) /
|
| 30 |
-
|
| 31 |
-
for i in range(
|
| 32 |
-
start, end = int(i * chunk_ms), int((i + 1) * chunk_ms)
|
| 33 |
-
chunk = audio[start:end]
|
| 34 |
fn = f"chunk_{i+1}.wav"
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
return
|
| 38 |
|
| 39 |
def transcribe_core(path, model="whisper-1"):
|
| 40 |
-
if path
|
| 41 |
-
|
| 42 |
try:
|
| 43 |
-
shutil.copy(path,
|
| 44 |
-
path =
|
| 45 |
-
print("🔧 已自動修正 mp4 → m4a")
|
| 46 |
except Exception as e:
|
| 47 |
print(f"⚠️ mp4→m4a 轉檔失敗:{e}")
|
| 48 |
|
| 49 |
chunks = split_audio_if_needed(path)
|
| 50 |
-
|
| 51 |
for f in chunks:
|
| 52 |
with open(f, "rb") as af:
|
| 53 |
-
res = client.audio.transcriptions.create(
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
response_format="text"
|
| 57 |
-
)
|
| 58 |
-
raw_parts.append(res)
|
| 59 |
-
full_raw = "\n".join(raw_parts)
|
| 60 |
|
| 61 |
-
# 簡轉繁
|
| 62 |
conv_prompt = (
|
| 63 |
"請將以下內容完整轉換為「繁體中文(台灣用語)」:\n"
|
| 64 |
-
"規則:1) 僅做簡→繁字形轉換;2) 不要意譯或改寫;3) 不要添加任何前後綴。\n"
|
| 65 |
-
"-----\n" + full_raw
|
| 66 |
)
|
| 67 |
-
|
| 68 |
model="gpt-4o-mini",
|
| 69 |
messages=[
|
| 70 |
-
{"role": "system", "content": "
|
| 71 |
{"role": "user", "content": conv_prompt}
|
| 72 |
],
|
| 73 |
temperature=0.0,
|
| 74 |
-
)
|
| 75 |
-
full_trad = trad_resp.choices[0].message.content.strip()
|
| 76 |
|
| 77 |
-
# 摘要
|
| 78 |
sum_prompt = (
|
| 79 |
-
"
|
| 80 |
-
+
|
| 81 |
)
|
| 82 |
-
|
| 83 |
model="gpt-4o-mini",
|
| 84 |
messages=[
|
| 85 |
{"role": "system", "content": "你是一位精準且嚴格使用台灣繁體中文的摘要助手。"},
|
| 86 |
{"role": "user", "content": sum_prompt}
|
| 87 |
],
|
| 88 |
temperature=0.2,
|
| 89 |
-
)
|
| 90 |
-
|
| 91 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
|
| 93 |
# ======================================================
|
| 94 |
# 💬 Gradio 介面
|
| 95 |
# ======================================================
|
| 96 |
def transcribe_with_password(password, file):
|
| 97 |
if password.strip() != PASSWORD:
|
| 98 |
-
|
| 99 |
if not file:
|
| 100 |
return "⚠️ 未選擇檔案", "", ""
|
| 101 |
text, summary = transcribe_core(file.name)
|
|
@@ -114,7 +121,11 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 114 |
# ======================================================
|
| 115 |
# 🚀 啟動
|
| 116 |
# ======================================================
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
if __name__ == "__main__":
|
| 118 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|
| 119 |
-
else:
|
| 120 |
-
demo.launch()
|
|
|
|
| 4 |
from pydub import AudioSegment
|
| 5 |
from openai import OpenAI
|
| 6 |
import gradio as gr
|
| 7 |
+
from fastapi import FastAPI, UploadFile, File, Form
|
| 8 |
+
from threading import Thread
|
| 9 |
+
import uvicorn
|
| 10 |
|
| 11 |
# ======================================================
|
| 12 |
# 🔐 設定區
|
| 13 |
# ======================================================
|
| 14 |
+
PASSWORD = os.getenv("APP_PASSWORD", "chou")
|
| 15 |
MAX_SIZE = 25 * 1024 * 1024
|
| 16 |
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
| 17 |
|
|
|
|
| 27 |
if size <= MAX_SIZE:
|
| 28 |
return [path]
|
| 29 |
audio = AudioSegment.from_file(path)
|
| 30 |
+
n = int(size / MAX_SIZE) + 1
|
| 31 |
+
chunk_ms = len(audio) / n
|
| 32 |
+
parts = []
|
| 33 |
+
for i in range(n):
|
|
|
|
|
|
|
| 34 |
fn = f"chunk_{i+1}.wav"
|
| 35 |
+
audio[int(i * chunk_ms):int((i + 1) * chunk_ms)].export(fn, format="wav")
|
| 36 |
+
parts.append(fn)
|
| 37 |
+
return parts
|
| 38 |
|
| 39 |
def transcribe_core(path, model="whisper-1"):
|
| 40 |
+
if path.lower().endswith(".mp4"):
|
| 41 |
+
fixed = path[:-4] + ".m4a"
|
| 42 |
try:
|
| 43 |
+
shutil.copy(path, fixed)
|
| 44 |
+
path = fixed
|
|
|
|
| 45 |
except Exception as e:
|
| 46 |
print(f"⚠️ mp4→m4a 轉檔失敗:{e}")
|
| 47 |
|
| 48 |
chunks = split_audio_if_needed(path)
|
| 49 |
+
txts = []
|
| 50 |
for f in chunks:
|
| 51 |
with open(f, "rb") as af:
|
| 52 |
+
res = client.audio.transcriptions.create(model=model, file=af, response_format="text")
|
| 53 |
+
txts.append(res)
|
| 54 |
+
full_raw = "\n".join(txts)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
|
|
|
| 56 |
conv_prompt = (
|
| 57 |
"請將以下內容完整轉換為「繁體中文(台灣用語)」:\n"
|
| 58 |
+
"規則:1) 僅做簡→繁字形轉換;2) 不要意譯或改寫;3) 不要添加任何前後綴。\n-----\n" + full_raw
|
|
|
|
| 59 |
)
|
| 60 |
+
trad = client.chat.completions.create(
|
| 61 |
model="gpt-4o-mini",
|
| 62 |
messages=[
|
| 63 |
+
{"role": "system", "content": "你是嚴格的繁體中文轉換器。"},
|
| 64 |
{"role": "user", "content": conv_prompt}
|
| 65 |
],
|
| 66 |
temperature=0.0,
|
| 67 |
+
).choices[0].message.content.strip()
|
|
|
|
| 68 |
|
|
|
|
| 69 |
sum_prompt = (
|
| 70 |
+
"請用台灣繁體中文撰寫摘要。若內容資訊多,可條列出重點;"
|
| 71 |
+
"若內容簡短,請用一句話概述即可。\n\n" + trad
|
| 72 |
)
|
| 73 |
+
summ = client.chat.completions.create(
|
| 74 |
model="gpt-4o-mini",
|
| 75 |
messages=[
|
| 76 |
{"role": "system", "content": "你是一位精準且嚴格使用台灣繁體中文的摘要助手。"},
|
| 77 |
{"role": "user", "content": sum_prompt}
|
| 78 |
],
|
| 79 |
temperature=0.2,
|
| 80 |
+
).choices[0].message.content.strip()
|
| 81 |
+
|
| 82 |
+
return trad, summ
|
| 83 |
+
|
| 84 |
+
# ======================================================
|
| 85 |
+
# 🌐 FastAPI for 捷徑
|
| 86 |
+
# ======================================================
|
| 87 |
+
api_app = FastAPI()
|
| 88 |
+
|
| 89 |
+
@api_app.post("/api/transcribe")
|
| 90 |
+
async def api_transcribe(file: UploadFile = File(...), token: str = Form(...)):
|
| 91 |
+
if token != PASSWORD:
|
| 92 |
+
return {"error": "Invalid token"}
|
| 93 |
+
temp = file.filename
|
| 94 |
+
with open(temp, "wb") as f:
|
| 95 |
+
f.write(await file.read())
|
| 96 |
+
text, summary = transcribe_core(temp)
|
| 97 |
+
os.remove(temp)
|
| 98 |
+
return {"text": text, "summary": summary}
|
| 99 |
|
| 100 |
# ======================================================
|
| 101 |
# 💬 Gradio 介面
|
| 102 |
# ======================================================
|
| 103 |
def transcribe_with_password(password, file):
|
| 104 |
if password.strip() != PASSWORD:
|
| 105 |
+
return "❌ 密碼錯誤", "", ""
|
| 106 |
if not file:
|
| 107 |
return "⚠️ 未選擇檔案", "", ""
|
| 108 |
text, summary = transcribe_core(file.name)
|
|
|
|
| 121 |
# ======================================================
|
| 122 |
# 🚀 啟動
|
| 123 |
# ======================================================
|
| 124 |
+
def run_api():
|
| 125 |
+
uvicorn.run(api_app, host="0.0.0.0", port=7861)
|
| 126 |
+
|
| 127 |
+
Thread(target=run_api, daemon=True).start()
|
| 128 |
+
app = demo # ✅ Hugging Face 主入口使用 Gradio
|
| 129 |
+
|
| 130 |
if __name__ == "__main__":
|
| 131 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
|
|