Update app.py

app.py

@@ -15,13 +15,13 @@ import torch
 import torchaudio as ta
 import gradio as gr
 
-# Optimasi Core CPU shared Hugging Face agar fokus
+# Optimasi Core CPU shared Hugging Face agar fokus dan stabil
 torch.set_num_threads(2)
 torch.set_num_interop_threads(2)
 torch.set_grad_enabled(False)
 torch.cuda.is_available = lambda: False  # noqa: E731
 
-# Override global torch load untuk keamanan CPU
+# Override global torch load untuk keamanan jalur CPU
 _original_torch_load = torch.load
 def _torch_load_cpu(*args, **kwargs):
     kwargs["map_location"] = torch.device("cpu")
@@ -41,13 +41,13 @@ MAX_CHUNKS = int(os.getenv("MAX_CHUNKS", "12"))
 PAUSE_SECONDS = float(os.getenv("PAUSE_SECONDS", "0.15"))
 DOWNLOAD_TIMEOUT = int(os.getenv("DOWNLOAD_TIMEOUT", "90"))
 
-# Jalur untuk suara default bawaan sistem (Taruh file suara 1:15 kamu di folder assets Space ini)
+# Jalur untuk suara default bawaan sistem (Simpan file 1:15 kamu di folder assets)
 ROOT_DIR = Path(__file__).parent
 ASSETS_DIR = ROOT_DIR / "assets"
 ASSETS_DIR.mkdir(exist_ok=True)
 DEFAULT_SPEAKER_PATH = ASSETS_DIR / "default_speaker.wav"
 
-# Global Cache untuk mengunci embedding suara agar tidak dihitung ulang setiap request
+# Global Cache untuk mengunci embedding karakteristik suara ke dalam RAM
 CACHED_EMBEDDINGS = {}
 
 from chatterbox.tts import ChatterboxTTS
@@ -98,7 +98,7 @@ def _resolve_audio_input(audio_file, audio_url: str):
         except Exception:
             pass
             
-    # KUNCI UTAMA: Jika tidak ada file/url yang dikirim, otomatis alihkan ke Default Voice bawaan sistem
+    # Jika parameter audio kosong, otomatis gunakan file suara default 1:15 di folder assets
     if DEFAULT_SPEAKER_PATH.exists():
         return str(DEFAULT_SPEAKER_PATH)
         
@@ -147,7 +147,7 @@ def _split_text_safely(text: str, max_chars: int = MAX_CHARS_PER_CHUNK):
     return chunks
 
 # =====================================================================
-# ENGINE INFERENCE DENGAN DUAL-SPEAKER SELECTION & EMBEDDING CACHE
+# ENGINE UTAMA: SUDAH FIX TYPEERROR ARGUMEN & CACHE EMBEDDING
 # =====================================================================
 def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(track_tqdm=False)):
     global CACHED_EMBEDDINGS
@@ -156,10 +156,10 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
         if not raw_text:
             raise gr.Error("Text prompt tidak boleh kosong.")
 
-        # Tentukan file suara yang dipakai (Custom atau Default Lokal)
+        # Ambil jalur audio (Custom atau Default)
         prompt_path = _resolve_audio_input(audio_file, audio_url)
         if not prompt_path:
-            raise gr.Error("Suara default tidak ditemukan di server. Harap upload sampel audio.")
+            raise gr.Error("Suara acuan tidak ditemukan. Sediakan file audio default atau upload sampel.")
 
         chunks = _split_text_safely(raw_text, max_chars=MAX_CHARS_PER_CHUNK)
         chunks = [ch for ch in chunks if re.search(r'[a-zA-Z0-9]', ch)]
@@ -168,9 +168,9 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
         sr = getattr(model, "sr", 24000)
         torch.manual_seed(42)
 
-        # Caching Dinamis berdasarkan path audio agar multi-user/multi-voice tetap aman
+        # Proses pembuatan cache embedding jika file audio terdeteksi baru
         if prompt_path not in CACHED_EMBEDDINGS:
-            progress(0.0, desc="Mengekstrak dan mengunci embedding karakteristik suara ke RAM...")
+            progress(0.0, desc="Mengekstrak karakteristik gelombang audio ke RAM (Hanya sekali)...")
             
             if hasattr(model, "extract_conditioning") or hasattr(model, "get_speaker_embedding"):
                 extract_fn = getattr(model, "extract_conditioning", getattr(model, "get_speaker_embedding", None))
@@ -178,7 +178,7 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
             else:
                 CACHED_EMBEDDINGS[prompt_path] = prompt_path
                 
-            print(f"[CACHE] Karakteristik untuk {prompt_path} berhasil dikunci.")
+            print(f"[CACHE] Fitur karakteristik untuk {prompt_path} berhasil dikunci di RAM.")
 
         speaker_embedding = CACHED_EMBEDDINGS[prompt_path]
 
@@ -186,6 +186,7 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
         pause = torch.zeros(1, int(sr * PAUSE_SECONDS))
         total = len(chunks)
         
+        # Bedah blueprint fungsi model asli untuk memvalidasi parameter masuk
         sig = inspect.signature(model.generate)
         params = sig.parameters
 
@@ -197,21 +198,31 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
                 if not re.search(r"[.!?…]$", ch):
                     ch += "."
 
-                kwargs = {
-                    "temperature": 0.05,
-                    "top_p": 0.7,
-                    "exaggeration": 0.25,
-                    "cfg_weight": 0.3,
-                    "max_new_tokens": 260
-                }
+                # Saring parameter agar hanya memasukkan key yang dikenal oleh library model
+                kwargs = {}
                 
                 if "audio_prompt_path" in params:
                     kwargs["audio_prompt_path"] = speaker_embedding
                 elif "speaker_embedding" in params:
                     kwargs["speaker_embedding"] = speaker_embedding
-                else:
-                    kwargs["audio_prompt_path"] = prompt_path
-
+                elif "prompt" in params and len(params) > 1:
+                    list_keys = list(params.keys())
+                    if len(list_keys) > 1:
+                        kwargs[list_keys[1]] = speaker_embedding
+                
+                # Parameter opsional tambahan (Hanya disuntikkan jika disupport oleh versi model)
+                if "temperature" in params:
+                    kwargs["temperature"] = 0.05
+                if "top_p" in params:
+                    kwargs["top_p"] = 0.7
+                if "exaggeration" in params:
+                    kwargs["exaggeration"] = 0.25
+                if "cfg_weight" in params:
+                    kwargs["cfg_weight"] = 0.3
+                if "max_new_tokens" in params:
+                    kwargs["max_new_tokens"] = 260
+
+                # Eksekusi aman bebas dari bug TypeError
                 try:
                     wav = model.generate(ch, **kwargs)
                 except TypeError:
@@ -219,6 +230,7 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
                         kwargs["text"] = ch
                         wav = model.generate(**kwargs)
                     else:
+                        # Jalur teraman jika custom kwargs ditolak total oleh internal model
                         wav = model.generate(ch)
 
                 if wav.dim() == 1:
@@ -230,12 +242,13 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
         if wav_parts:
             wav_parts = wav_parts[:-1]
 
-        progress(0.95, desc="Menggabungkan audio akhir...")
+        progress(0.95, desc="Menggabungkan kompilasi audio...")
         full_wav = torch.cat(wav_parts, dim=1)
 
         out_path = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
         ta.save(out_path, full_wav, sr)
 
+        # Bersihkan memori RAM kontainer secara agresif
         del wav_parts
         del full_wav
         gc.collect()
@@ -251,16 +264,16 @@ def clone_voice(text: str, audio_file, audio_url: str, progress=gr.Progress(trac
 # =====================================================================
 # INTERFACE DESIGN
 # =====================================================================
-with gr.Blocks(title="Chatterbox Dual Voice Hybrid Engine") as demo:
-    gr.Markdown("## EduScanner AI - Dual Voice Solution (Default & Custom)")
-    gr.Markdown("Otomatis menggunakan suara default jika input file dikosongkan.")
+with gr.Blocks(title="Chatterbox Dual Voice Safe Engine") as demo:
+    gr.Markdown("## EduScanner AI Voice Backend - Anti-Error Parameter Generation")
+    gr.Markdown("Kode ini menyaring argumen secara dinamis untuk mencegah error crash sistem.")
 
-    text_in = gr.Textbox(label="Teks Rangkuman", lines=8, placeholder="Masukkan teks di sini...")
-    wav_in = gr.Audio(label="Opsi Custom Voice (Kosongkan jika ingin pakai suara default Syahid)", type="filepath")
-    url_in = gr.Textbox(label="Opsi Link URL Audio Custom", placeholder="https://example.com/suara_lain.wav")
+    text_in = gr.Textbox(label="Teks Rangkuman Materi Kuliah", lines=8, placeholder="Ketik teks di sini...")
+    wav_in = gr.Audio(label="Opsi Custom Voice (Kosongkan jika ingin pakai suara default Syahid 1:15)", type="filepath")
+    url_in = gr.Textbox(label="Opsi URL File Audio Custom", placeholder="https://domain-kamu.com/audio.wav")
     
-    btn = gr.Button("Sintesis Audio", variant="primary")
-    out_audio = gr.Audio(label="Hasil Audio (WAV)", type="filepath")
+    btn = gr.Button("Sintesis Audio Kloning Suara", variant="primary")
+    out_audio = gr.Audio(label="Hasil Audio Akhir (WAV)", type="filepath")
 
     btn.click(
         fn=clone_voice,