Upgrade default Whisper model: small/medium → large-v3 for much better Slovenian/Slavic transcription accuracy

app/main.py

@@ -240,7 +240,7 @@ def process_job(job_id):
             # lang: če None ali 'auto', pusti analyze.py auto-detect
             if job.get("lang") and job["lang"] not in ("auto", ""):
                 cmd += ["--lang", job["lang"]]
-            cmd += ["--model", job.get("whisper_model", "small")]
+            cmd += ["--model", job.get("whisper_model", "large-v3")]
 
             proc = subprocess.run(cmd, capture_output=True, text=True)
             srt_from_claude = None  # Pot do SRT iz Claude-popravljenega transcript-a
@@ -334,7 +334,7 @@ def process_job(job_id):
             cmd += ["--lang", chosen_lang]
         if job.get("no_subs"):
             cmd += ["--no-subs"]
-        cmd += ["--model", job.get("whisper_model", "small")]
+        cmd += ["--model", job.get("whisper_model", "large-v3")]
 
         # DEBUG: zapiši natanko kakšen ukaz se izvede
         update_job(job_id, debug_clip_cmd=" ".join(cmd))
@@ -391,7 +391,7 @@ class YouTubeJobIn(BaseModel):
     duration: Optional[float] = 30
     no_subs: bool = False
     subtitle_style: str = "reels"
-    whisper_model: str = "small"
+    whisper_model: str = "large-v3"
     quality: str = "medium"
 
 
@@ -407,7 +407,7 @@ class StartJobIn(BaseModel):
     min_duration: Optional[float] = 20
     no_subs: bool = False
     subtitle_style: str = "reels"
-    whisper_model: str = "small"
+    whisper_model: str = "large-v3"
     quality: str = "medium"
 
 

scripts/analyze.py

@@ -592,7 +592,7 @@ def main():
     ap = argparse.ArgumentParser()
     ap.add_argument("video", help="Vhod video file")
     ap.add_argument("--lang", default=None, help="ISO 639-1 ali 'auto' (default: auto)")
-    ap.add_argument("--model", default="small", help="Whisper model")
+    ap.add_argument("--model", default="large-v3", help="Whisper model")
     ap.add_argument("--target-duration", type=float, default=30.0)
     ap.add_argument("--max-duration", type=float, default=45.0)
     ap.add_argument("--min-duration", type=float, default=20.0)

scripts/clip.py

@@ -112,7 +112,7 @@ def main():
                     help="Več klipov: '0:30-1:00,2:15-2:45'")
     ap.add_argument("--mode", default="track", choices=["track", "center", "blur"])
     ap.add_argument("--lang", default=None, help="sl, de, en, ... (privzeto auto)")
-    ap.add_argument("--model", default="small",
+    ap.add_argument("--model", default="large-v3",
                     choices=["tiny", "base", "small", "medium", "large-v3"])
     ap.add_argument("--style", default="reels", choices=["reels", "yellow", "minimal"])
     ap.add_argument("--no-subs", action="store_true")

scripts/subtitle.py

@@ -279,7 +279,7 @@ def main():
     ap.add_argument("input")
     ap.add_argument("output")
     ap.add_argument("--lang", default=None, help="Jezik (sl, de, en, ...) ali auto")
-    ap.add_argument("--model", default="small", choices=["tiny", "base", "small", "medium", "large-v3"])
+    ap.add_argument("--model", default="large-v3", choices=["tiny", "base", "small", "medium", "large-v3"])
     ap.add_argument("--style", default="reels", choices=list(SUBTITLE_STYLES.keys()))
     ap.add_argument("--keep-srt", action="store_true", help="Ohrani .srt poleg output")
     ap.add_argument("--srt", default=None, help="Že-pripravljen SRT (preskoči Whisper transkripcijo)")

templates/index.html

@@ -214,7 +214,7 @@
 
       <!-- Skrita polja: jezik in model sta avto. Vrednosti uporabljene v JS submit. -->
       <input type="hidden" id="lang" value="">
-      <input type="hidden" id="model" value="medium">
+      <input type="hidden" id="model" value="large-v3">
 
       <div style="font-size: 12px; color: var(--text-dim); margin-top: 8px;">
         🤖 Jezik: avtomatsko zaznan (Whisper, 3-sample voting) · Model: medium · LLM analiza: Claude