Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -9,106 +9,95 @@ from faster_whisper import WhisperModel
|
|
| 9 |
import tempfile
|
| 10 |
import time
|
| 11 |
import os
|
|
|
|
| 12 |
# import requests # Artık gerek yok
|
| 13 |
-
from transformers import pipeline
|
| 14 |
-
import torch
|
| 15 |
|
| 16 |
# ==================== CONFIG & MODELS ====================
|
| 17 |
|
| 18 |
-
# 1. WHISPER MODEL (Ses Deşifre)
|
| 19 |
MODEL_SIZE = "medium"
|
| 20 |
model = None
|
| 21 |
|
| 22 |
try:
|
| 23 |
-
print(f"
|
| 24 |
model = WhisperModel(MODEL_SIZE, device="cpu", compute_type="int8")
|
| 25 |
print("✅ Whisper Modeli Hazır!")
|
| 26 |
except Exception as e:
|
| 27 |
print(f"❌ Whisper Yükleme Hatası: {e}")
|
| 28 |
model = None
|
| 29 |
|
| 30 |
-
#
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
def load_translator():
|
| 44 |
-
global translator_pipe
|
| 45 |
-
if translator_pipe is None:
|
| 46 |
-
print("📥 Çeviri Modeli (NLLB-1.3B) yükleniyor... (Bu biraz zaman alabilir)")
|
| 47 |
-
# Daha büyük ve kaliteli model: 1.3B
|
| 48 |
-
translator_pipe = pipeline("translation", model="facebook/nllb-200-distilled-1.3B", device=-1)
|
| 49 |
-
print("✅ Çeviri Modeli Hazır!")
|
| 50 |
-
return translator_pipe
|
| 51 |
-
|
| 52 |
-
# ==================== AI FUNCTIONS (LOCAL) ====================
|
| 53 |
-
|
| 54 |
-
def summarize_locally(text: str, progress=gr.Progress()) -> str:
|
| 55 |
-
"""Yerel model (mT5) ile özetleme."""
|
| 56 |
-
if not text or "⚠️" in text: return "⚠️ Önce geçerli bir metin oluşturun."
|
| 57 |
|
| 58 |
-
|
| 59 |
-
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
try:
|
| 62 |
-
|
| 63 |
-
pipe = load_summarizer()
|
| 64 |
-
|
| 65 |
-
progress(0.5, desc="Metin özetleniyor...")
|
| 66 |
-
# Uyarıları önlemek için parametreler düzeltildi
|
| 67 |
-
# max_length yerine max_new_tokens kullanıyoruz
|
| 68 |
-
result = pipe(clean_text, max_new_tokens=128, min_length=10, do_sample=False)
|
| 69 |
|
| 70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
| 72 |
except Exception as e:
|
| 73 |
-
return f"❌
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
|
| 75 |
-
def
|
| 76 |
-
"""
|
| 77 |
if not text or "⚠️" in text: return "⚠️ Çevrilecek metin yok."
|
| 78 |
|
| 79 |
clean_text = text.split("───────────────────────────────────")[0].strip()
|
| 80 |
|
| 81 |
-
#
|
| 82 |
-
lang_map = {
|
| 83 |
-
|
| 84 |
-
"Almanca": "deu_Latn",
|
| 85 |
-
"Fransızca": "fra_Latn",
|
| 86 |
-
"Türkçe": "tur_Latn"
|
| 87 |
-
}
|
| 88 |
-
src_lang = "tur_Latn" # Varsayılan giriş Türkçe
|
| 89 |
-
tgt_lang = lang_map.get(target_language, "eng_Latn")
|
| 90 |
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
# NLLB pipeline kullanımı: src_lang ve tgt_lang belirtilmeli
|
| 97 |
-
# Tekrarı önlemek için paramatreler eklendi
|
| 98 |
-
result = pipe(
|
| 99 |
-
clean_text,
|
| 100 |
-
src_lang=src_lang,
|
| 101 |
-
tgt_lang=tgt_lang,
|
| 102 |
-
max_length=512,
|
| 103 |
-
repetition_penalty=1.2, # Tekrar cezası
|
| 104 |
-
no_repeat_ngram_size=3, # 3 kelimelik tekrarları yasakla
|
| 105 |
-
num_beams=3 # Rota arama kalitesini artır
|
| 106 |
-
)
|
| 107 |
-
|
| 108 |
-
return result[0]['translation_text']
|
| 109 |
-
|
| 110 |
-
except Exception as e:
|
| 111 |
-
return f"❌ Çeviri Hatası: {str(e)}"
|
| 112 |
|
| 113 |
|
| 114 |
# ==================== TRANSCRIPTION (WHISPER) ====================
|
|
@@ -166,48 +155,55 @@ def transcribe(audio_path: str, progress=gr.Progress()):
|
|
| 166 |
|
| 167 |
# ==================== UI (GRADIO) ====================
|
| 168 |
|
| 169 |
-
with gr.Blocks(title="Ses Deşifre Pro (
|
| 170 |
|
| 171 |
gr.HTML("""
|
| 172 |
<style>
|
| 173 |
footer { display: none !important; }
|
| 174 |
.gradio-container { max-width: 900px !important; margin: auto !important; }
|
| 175 |
</style>
|
| 176 |
-
<div style="text-align: center; padding: 30px; background: linear-gradient(135deg, #
|
| 177 |
-
<h1 style="font-size: 2.2rem; margin: 0;">🎙️ Ses Deşifre &
|
| 178 |
-
<p style="opacity: 0.9;"
|
| 179 |
</div>
|
| 180 |
""")
|
| 181 |
|
| 182 |
with gr.Row():
|
| 183 |
with gr.Column():
|
| 184 |
audio_input = gr.Audio(label="Ses Dosyası", type="filepath", sources=["upload", "microphone"])
|
| 185 |
-
submit_btn = gr.Button("🚀
|
| 186 |
|
| 187 |
with gr.Row():
|
| 188 |
with gr.Column():
|
| 189 |
-
output_text = gr.Textbox(label="
|
| 190 |
-
download_file = gr.File(label="
|
| 191 |
|
| 192 |
-
# ---
|
| 193 |
-
gr.HTML("<h3 style='margin-top: 20px; border-bottom: 1px solid #ddd; padding-bottom: 10px;'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
|
| 195 |
with gr.Tabs():
|
| 196 |
-
with gr.TabItem("✨ Özetle
|
| 197 |
-
summary_btn = gr.Button("📝
|
| 198 |
summary_output = gr.Textbox(label="Özet Sonucu", lines=6)
|
| 199 |
|
| 200 |
-
with gr.TabItem("🌍 Çevir
|
| 201 |
with gr.Row():
|
| 202 |
target_lang = gr.Dropdown(["İngilizce", "Almanca", "Fransızca"], label="Hedef Dil", value="İngilizce")
|
| 203 |
-
translate_btn = gr.Button("A Çevir")
|
| 204 |
translate_output = gr.Textbox(label="Çeviri Sonucu", lines=6)
|
| 205 |
|
| 206 |
# --- BAĞLANTILAR ---
|
| 207 |
submit_btn.click(transcribe, inputs=[audio_input], outputs=[output_text, download_file])
|
| 208 |
|
| 209 |
-
summary_btn.click(
|
| 210 |
-
translate_btn.click(
|
| 211 |
|
| 212 |
if __name__ == "__main__":
|
| 213 |
demo.launch(share=False)
|
|
|
|
| 9 |
import tempfile
|
| 10 |
import time
|
| 11 |
import os
|
| 12 |
+
import requests
|
| 13 |
# import requests # Artık gerek yok
|
| 14 |
+
# from transformers import pipeline
|
| 15 |
+
# import torch
|
| 16 |
|
| 17 |
# ==================== CONFIG & MODELS ====================
|
| 18 |
|
| 19 |
+
# 1. WHISPER MODEL (Ses Deşifre - CPU/Local)
|
| 20 |
MODEL_SIZE = "medium"
|
| 21 |
model = None
|
| 22 |
|
| 23 |
try:
|
| 24 |
+
print(f"📥 Whisper {MODEL_SIZE} modeli yükleniyor...")
|
| 25 |
model = WhisperModel(MODEL_SIZE, device="cpu", compute_type="int8")
|
| 26 |
print("✅ Whisper Modeli Hazır!")
|
| 27 |
except Exception as e:
|
| 28 |
print(f"❌ Whisper Yükleme Hatası: {e}")
|
| 29 |
model = None
|
| 30 |
|
| 31 |
+
# ==================== AI API FUNCTIONS (Hugging Face) ====================
|
| 32 |
+
|
| 33 |
+
def call_huggingface_api(prompt, api_key, model_id):
|
| 34 |
+
"""
|
| 35 |
+
Hugging Face Serverless Inference API (Legacy Endpoint).
|
| 36 |
+
Standart 'Read' token ile çalışır.
|
| 37 |
+
"""
|
| 38 |
+
if not api_key: return "⚠️ HF Token girilmedi."
|
| 39 |
+
if not api_key.startswith("hf_"): return "⚠️ Token 'hf_' ile başlamalıdır."
|
| 40 |
+
|
| 41 |
+
url = f"https://api-inference.huggingface.co/models/{model_id}"
|
| 42 |
+
headers = {"Authorization": f"Bearer {api_key}"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
+
# Zephyr/Mistral Prompt Formatı
|
| 45 |
+
formatted_prompt = f"<|system|>\nSen yardımsever bir asistansın.\n<|user|>\n{prompt}\n<|assistant|>\n"
|
| 46 |
|
| 47 |
+
payload = {
|
| 48 |
+
"inputs": formatted_prompt,
|
| 49 |
+
"parameters": {
|
| 50 |
+
"max_new_tokens": 512,
|
| 51 |
+
"return_full_text": False,
|
| 52 |
+
"temperature": 0.3
|
| 53 |
+
}
|
| 54 |
+
}
|
| 55 |
+
|
| 56 |
try:
|
| 57 |
+
response = requests.post(url, headers=headers, json=payload, timeout=60)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
+
if response.status_code == 200:
|
| 60 |
+
result = response.json()
|
| 61 |
+
if isinstance(result, list) and len(result) > 0 and "generated_text" in result[0]:
|
| 62 |
+
return result[0]["generated_text"].strip()
|
| 63 |
+
elif isinstance(result, dict) and "generated_text" in result:
|
| 64 |
+
return result["generated_text"].strip()
|
| 65 |
+
return f"❌ API Beklenmedik Yanıt: {result}"
|
| 66 |
+
|
| 67 |
+
elif response.status_code == 503:
|
| 68 |
+
return "⚠️ Model şu an yükleniyor (Cold Boot). 30 saniye sonra tekrar deneyin."
|
| 69 |
+
|
| 70 |
+
else:
|
| 71 |
+
return f"❌ API Hatası ({response.status_code}): {response.text}"
|
| 72 |
|
| 73 |
except Exception as e:
|
| 74 |
+
return f"❌ Bağlantı Hatası: {str(e)}"
|
| 75 |
+
|
| 76 |
+
def summarize_with_api(text: str, api_key: str) -> str:
|
| 77 |
+
"""Zephyr-7B kullanarak özetler."""
|
| 78 |
+
if not text or "⚠️" in text: return "⚠️ Özetlenecek metin yok."
|
| 79 |
+
|
| 80 |
+
clean_text = text.split("───────────────────────────────────")[0].strip()
|
| 81 |
+
prompt = f"Aşağıdaki metni Türkçe olarak maddeler halinde özetle:\n\n{clean_text}"
|
| 82 |
+
|
| 83 |
+
# Model: Zephyr 7B Beta (Genelde çok stabil ve ücretsiz)
|
| 84 |
+
return call_huggingface_api(prompt, api_key, "HuggingFaceH4/zephyr-7b-beta")
|
| 85 |
|
| 86 |
+
def translate_with_api(text: str, target_language: str, api_key: str) -> str:
|
| 87 |
+
"""Facebook NLLB kullanarak çevirir (API üzerinden)."""
|
| 88 |
if not text or "⚠️" in text: return "⚠️ Çevrilecek metin yok."
|
| 89 |
|
| 90 |
clean_text = text.split("───────────────────────────────────")[0].strip()
|
| 91 |
|
| 92 |
+
# Dil haritası
|
| 93 |
+
lang_map = {"İngilizce": "English", "Almanca": "German", "Fransızca": "French", "Türkçe": "Turkish"}
|
| 94 |
+
tgt = lang_map.get(target_language, "English")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
+
prompt = f"Translate the following text to {tgt}. Only provide the translation, no extra text.\n\nText:\n{clean_text}"
|
| 97 |
+
|
| 98 |
+
# Model: Zephyr çeviri için de iyidir, NLLB API'si bazen kararsız olabilir.
|
| 99 |
+
# Yine de NLLB deneyebiliriz ama Zephyr daha genel amaçlı ve sağlamdır.
|
| 100 |
+
return call_huggingface_api(prompt, api_key, "HuggingFaceH4/zephyr-7b-beta")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
|
| 102 |
|
| 103 |
# ==================== TRANSCRIPTION (WHISPER) ====================
|
|
|
|
| 155 |
|
| 156 |
# ==================== UI (GRADIO) ====================
|
| 157 |
|
| 158 |
+
with gr.Blocks(title="Ses Deşifre Pro (Whisper + HF API)") as demo:
|
| 159 |
|
| 160 |
gr.HTML("""
|
| 161 |
<style>
|
| 162 |
footer { display: none !important; }
|
| 163 |
.gradio-container { max-width: 900px !important; margin: auto !important; }
|
| 164 |
</style>
|
| 165 |
+
<div style="text-align: center; padding: 30px; background: linear-gradient(135deg, #6366f1 0%, #a855f7 100%); border-radius: 20px; margin-bottom: 20px; color: white;">
|
| 166 |
+
<h1 style="font-size: 2.2rem; margin: 0;">🎙️ Ses Deşifre & HF API</h1>
|
| 167 |
+
<p style="opacity: 0.9;">Whisper (Local) + Zephyr AI (Cloud) • Hızlı & Ücretsiz</p>
|
| 168 |
</div>
|
| 169 |
""")
|
| 170 |
|
| 171 |
with gr.Row():
|
| 172 |
with gr.Column():
|
| 173 |
audio_input = gr.Audio(label="Ses Dosyası", type="filepath", sources=["upload", "microphone"])
|
| 174 |
+
submit_btn = gr.Button("🚀 Deşifre Et", variant="primary", size="lg")
|
| 175 |
|
| 176 |
with gr.Row():
|
| 177 |
with gr.Column():
|
| 178 |
+
output_text = gr.Textbox(label="Metin", placeholder="Sonuçlar burada...", lines=10, interactive=False)
|
| 179 |
+
download_file = gr.File(label="İndir (.txt)")
|
| 180 |
|
| 181 |
+
# --- API GİRİŞİ ---
|
| 182 |
+
gr.HTML("<h3 style='margin-top: 20px; border-bottom: 1px solid #ddd; padding-bottom: 10px;'>☁️ Hugging Face API (Özet & Çeviri)</h3>")
|
| 183 |
+
|
| 184 |
+
with gr.Row():
|
| 185 |
+
api_key_input = gr.Textbox(
|
| 186 |
+
label="🔑 HF Token (Örn: hf_...)",
|
| 187 |
+
placeholder="Read yetkili token yapıştırın...",
|
| 188 |
+
type="password"
|
| 189 |
+
)
|
| 190 |
|
| 191 |
with gr.Tabs():
|
| 192 |
+
with gr.TabItem("✨ Özetle"):
|
| 193 |
+
summary_btn = gr.Button("📝 Token ile Özetle (Zephyr)")
|
| 194 |
summary_output = gr.Textbox(label="Özet Sonucu", lines=6)
|
| 195 |
|
| 196 |
+
with gr.TabItem("🌍 Çevir"):
|
| 197 |
with gr.Row():
|
| 198 |
target_lang = gr.Dropdown(["İngilizce", "Almanca", "Fransızca"], label="Hedef Dil", value="İngilizce")
|
| 199 |
+
translate_btn = gr.Button("A Çevir (Zephyr)")
|
| 200 |
translate_output = gr.Textbox(label="Çeviri Sonucu", lines=6)
|
| 201 |
|
| 202 |
# --- BAĞLANTILAR ---
|
| 203 |
submit_btn.click(transcribe, inputs=[audio_input], outputs=[output_text, download_file])
|
| 204 |
|
| 205 |
+
summary_btn.click(summarize_with_api, inputs=[output_text, api_key_input], outputs=summary_output)
|
| 206 |
+
translate_btn.click(translate_with_api, inputs=[output_text, target_lang, api_key_input], outputs=translate_output)
|
| 207 |
|
| 208 |
if __name__ == "__main__":
|
| 209 |
demo.launch(share=False)
|