Update app.py
Browse files
app.py
CHANGED
|
@@ -24,7 +24,7 @@ try:
|
|
| 24 |
"ivrit-ai/pyannote-speaker-diarization-3.1",
|
| 25 |
use_auth_token=HF_TOKEN,
|
| 26 |
)
|
| 27 |
-
pipeline.to(device)
|
| 28 |
print("✅ Model loaded successfully!")
|
| 29 |
except Exception as e:
|
| 30 |
print(f"❌ Failed to load model: {e}")
|
|
@@ -65,7 +65,6 @@ def estimate_duration(file_path: str) -> float:
|
|
| 65 |
"""אומדן אורך קובץ בדקות"""
|
| 66 |
try:
|
| 67 |
file_size_mb = os.path.getsize(file_path) / (1024 * 1024)
|
| 68 |
-
# אומדן: ~2MB לדקה (ממוצע)
|
| 69 |
return file_size_mb / 2.0
|
| 70 |
except Exception:
|
| 71 |
return 0
|
|
@@ -128,39 +127,32 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 128 |
wav_path = None
|
| 129 |
|
| 130 |
try:
|
| 131 |
-
# קריאת הקובץ
|
| 132 |
content = await file.read()
|
| 133 |
file_size_mb = len(content) / (1024 * 1024)
|
| 134 |
|
| 135 |
-
# בדיקת גודל
|
| 136 |
if file_size_mb > MAX_FILE_SIZE_MB:
|
| 137 |
raise HTTPException(
|
| 138 |
status_code=400,
|
| 139 |
detail=f"File too large: {file_size_mb:.1f}MB (max: {MAX_FILE_SIZE_MB}MB)"
|
| 140 |
)
|
| 141 |
|
| 142 |
-
# בדיקת תור
|
| 143 |
if active_requests >= MAX_CONCURRENT_REQUESTS:
|
| 144 |
raise HTTPException(
|
| 145 |
status_code=503,
|
| 146 |
detail=f"Server busy ({active_requests}/{MAX_CONCURRENT_REQUESTS} active). Try again later."
|
| 147 |
)
|
| 148 |
|
| 149 |
-
# עיבוד
|
| 150 |
async with processing_semaphore:
|
| 151 |
active_requests += 1
|
| 152 |
|
| 153 |
try:
|
| 154 |
-
# שמירה זמנית
|
| 155 |
suffix = Path(file.filename).suffix or ".tmp"
|
| 156 |
with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
|
| 157 |
tmp.write(content)
|
| 158 |
tmp_path = tmp.name
|
| 159 |
|
| 160 |
-
# המרה ל-WAV
|
| 161 |
wav_path = ensure_wav_16k_mono(tmp_path)
|
| 162 |
|
| 163 |
-
# אומדן אורך
|
| 164 |
duration = estimate_duration(wav_path)
|
| 165 |
if duration > MAX_DURATION_MINUTES:
|
| 166 |
raise HTTPException(
|
|
@@ -168,7 +160,6 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 168 |
detail=f"File too long: ~{duration:.1f} min (max: {MAX_DURATION_MINUTES} min)"
|
| 169 |
)
|
| 170 |
|
| 171 |
-
# זיהוי דוברים
|
| 172 |
print(f"🎤 Processing: {file.filename} ({file_size_mb:.1f}MB)")
|
| 173 |
start_time = datetime.now()
|
| 174 |
|
|
@@ -177,7 +168,6 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 177 |
processing_time = (datetime.now() - start_time).total_seconds()
|
| 178 |
print(f"✅ Done in {processing_time:.1f}s")
|
| 179 |
|
| 180 |
-
# בניית תוצאות
|
| 181 |
segments = []
|
| 182 |
last_segment = None
|
| 183 |
|
|
@@ -186,7 +176,6 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 186 |
end = round(float(segment.end), 3)
|
| 187 |
speaker_id = str(speaker)
|
| 188 |
|
| 189 |
-
# איחוד מקטעים צמודים של אותו דובר
|
| 190 |
if (last_segment and
|
| 191 |
last_segment["speaker"] == speaker_id and
|
| 192 |
abs(start - last_segment["end"]) < 0.1):
|
|
@@ -203,7 +192,6 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 203 |
if last_segment:
|
| 204 |
segments.append(last_segment)
|
| 205 |
|
| 206 |
-
# סינון מקטעים קצרים
|
| 207 |
segments = [s for s in segments if s["end"] - s["start"] >= 0.2]
|
| 208 |
|
| 209 |
return JSONResponse({
|
|
@@ -224,7 +212,6 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 224 |
print(f"❌ Error: {str(e)}")
|
| 225 |
raise HTTPException(status_code=500, detail=f"Processing error: {str(e)}")
|
| 226 |
finally:
|
| 227 |
-
# ניקוי
|
| 228 |
for path in [tmp_path, wav_path]:
|
| 229 |
if path and os.path.exists(path):
|
| 230 |
try:
|
|
@@ -235,4 +222,21 @@ async def diarize(file: UploadFile = File(...)):
|
|
| 235 |
|
| 236 |
if __name__ == "__main__":
|
| 237 |
import uvicorn
|
| 238 |
-
uvicorn.run(app, host="0.0.0.0", port=7860)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
"ivrit-ai/pyannote-speaker-diarization-3.1",
|
| 25 |
use_auth_token=HF_TOKEN,
|
| 26 |
)
|
| 27 |
+
pipeline.to(torch.device(device)) # ✅ תיקון כאן!
|
| 28 |
print("✅ Model loaded successfully!")
|
| 29 |
except Exception as e:
|
| 30 |
print(f"❌ Failed to load model: {e}")
|
|
|
|
| 65 |
"""אומדן אורך קובץ בדקות"""
|
| 66 |
try:
|
| 67 |
file_size_mb = os.path.getsize(file_path) / (1024 * 1024)
|
|
|
|
| 68 |
return file_size_mb / 2.0
|
| 69 |
except Exception:
|
| 70 |
return 0
|
|
|
|
| 127 |
wav_path = None
|
| 128 |
|
| 129 |
try:
|
|
|
|
| 130 |
content = await file.read()
|
| 131 |
file_size_mb = len(content) / (1024 * 1024)
|
| 132 |
|
|
|
|
| 133 |
if file_size_mb > MAX_FILE_SIZE_MB:
|
| 134 |
raise HTTPException(
|
| 135 |
status_code=400,
|
| 136 |
detail=f"File too large: {file_size_mb:.1f}MB (max: {MAX_FILE_SIZE_MB}MB)"
|
| 137 |
)
|
| 138 |
|
|
|
|
| 139 |
if active_requests >= MAX_CONCURRENT_REQUESTS:
|
| 140 |
raise HTTPException(
|
| 141 |
status_code=503,
|
| 142 |
detail=f"Server busy ({active_requests}/{MAX_CONCURRENT_REQUESTS} active). Try again later."
|
| 143 |
)
|
| 144 |
|
|
|
|
| 145 |
async with processing_semaphore:
|
| 146 |
active_requests += 1
|
| 147 |
|
| 148 |
try:
|
|
|
|
| 149 |
suffix = Path(file.filename).suffix or ".tmp"
|
| 150 |
with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
|
| 151 |
tmp.write(content)
|
| 152 |
tmp_path = tmp.name
|
| 153 |
|
|
|
|
| 154 |
wav_path = ensure_wav_16k_mono(tmp_path)
|
| 155 |
|
|
|
|
| 156 |
duration = estimate_duration(wav_path)
|
| 157 |
if duration > MAX_DURATION_MINUTES:
|
| 158 |
raise HTTPException(
|
|
|
|
| 160 |
detail=f"File too long: ~{duration:.1f} min (max: {MAX_DURATION_MINUTES} min)"
|
| 161 |
)
|
| 162 |
|
|
|
|
| 163 |
print(f"🎤 Processing: {file.filename} ({file_size_mb:.1f}MB)")
|
| 164 |
start_time = datetime.now()
|
| 165 |
|
|
|
|
| 168 |
processing_time = (datetime.now() - start_time).total_seconds()
|
| 169 |
print(f"✅ Done in {processing_time:.1f}s")
|
| 170 |
|
|
|
|
| 171 |
segments = []
|
| 172 |
last_segment = None
|
| 173 |
|
|
|
|
| 176 |
end = round(float(segment.end), 3)
|
| 177 |
speaker_id = str(speaker)
|
| 178 |
|
|
|
|
| 179 |
if (last_segment and
|
| 180 |
last_segment["speaker"] == speaker_id and
|
| 181 |
abs(start - last_segment["end"]) < 0.1):
|
|
|
|
| 192 |
if last_segment:
|
| 193 |
segments.append(last_segment)
|
| 194 |
|
|
|
|
| 195 |
segments = [s for s in segments if s["end"] - s["start"] >= 0.2]
|
| 196 |
|
| 197 |
return JSONResponse({
|
|
|
|
| 212 |
print(f"❌ Error: {str(e)}")
|
| 213 |
raise HTTPException(status_code=500, detail=f"Processing error: {str(e)}")
|
| 214 |
finally:
|
|
|
|
| 215 |
for path in [tmp_path, wav_path]:
|
| 216 |
if path and os.path.exists(path):
|
| 217 |
try:
|
|
|
|
| 222 |
|
| 223 |
if __name__ == "__main__":
|
| 224 |
import uvicorn
|
| 225 |
+
uvicorn.run(app, host="0.0.0.0", port=7860)
|
| 226 |
+
```
|
| 227 |
+
|
| 228 |
+
---
|
| 229 |
+
|
| 230 |
+
## 🚀 עכשיו זה אמור לעבוד!
|
| 231 |
+
|
| 232 |
+
1. **עדכן את `app.py`** ב-Space
|
| 233 |
+
2. שמור
|
| 234 |
+
3. המתן לבנייה (~1-2 דקות - כי יש cache)
|
| 235 |
+
4. בדוק `/health`
|
| 236 |
+
|
| 237 |
+
אחרי התיקון הזה, אמור לראות בלוגים:
|
| 238 |
+
```
|
| 239 |
+
🚀 Loading model on cpu...
|
| 240 |
+
✅ Model loaded successfully!
|
| 241 |
+
INFO: Started server process
|
| 242 |
+
INFO: Uvicorn running on http://0.0.0.0:7860
|