feat: tích hợp giọng OmniVoice mới (fun_male/story_male/meme_male/calm_fun_male) + Cloud API https://hoangtaiii-omnivoice.hf.space — đồng bộ 100% với D:\omnivoice - web

app.py

@@ -316,11 +316,18 @@ with gr.Blocks(title="Trung Sáng Việt Cloud Studio") as demo:
 
             voice_input = gr.Dropdown(
                 choices=[
-                    ("🎙️ Nam Minh (Nam - Trầm ấm, chuyên nghiệp, phim tài liệu)", "vi-VN-NamMinhNeural"),
-                    ("🎙️ Hoài My (Nữ - Truyền cảm, ngọt ngào, review thời trang)", "vi-VN-HoaiMyNeural")
+                    ("🎙️ Nam Minh (Edge-TTS - Trầm ấm, chuyên nghiệp)", "vi-VN-NamMinhNeural"),
+                    ("🎙️ Hoài My (Edge-TTS - Truyền cảm, ngọt ngào)", "vi-VN-HoaiMyNeural"),
+                    ("🔥 Meme Male - Gasp (OmniVoice Cloud - viral, trẻ, high pitch)", "meme_male_gasp"),
+                    ("😎 Meme Male - Normal (OmniVoice Cloud)", "meme_male"),
+                    ("⚡ Fun Male - High Pitch (OmniVoice Cloud - vui nhộn)", "fun_male"),
+                    ("📖 Story Male - Kể chuyện (OmniVoice Cloud)", "story_male"),
+                    ("😌 Calm Fun Male - Trầm ấm vui (OmniVoice Cloud - low pitch)", "calm_fun_male"),
+                    ("🎭 Meme Excited - Hào hứng (OmniVoice Cloud)", "meme_male_excited"),
+                    ("💪 Meme Confident - Tự tin (OmniVoice Cloud)", "meme_male_confident"),
                 ],
-                value="vi-VN-NamMinhNeural",
-                label="Giọng đọc tiếng Việt (Microsoft Edge-TTS)"
+                value="meme_male_gasp",
+                label="Giọng đọc tiếng Việt (Edge-TTS / OmniVoice Cloud https://hoangtaiii-omnivoice.hf.space)"
             )
 
             source_lang_input = gr.Dropdown(

app/core/cloud_tts.py

@@ -21,12 +21,55 @@ from typing import List, Dict, Optional, Callable
 
 from app.core.vietnamese_text_normalizer import VietnameseTextNormalizer
 
+# ── OmniVoice presets — sync với D:\omnivoice - web\app.py ──
+OMNIVOICE_PRESETS = {
+    "fun_male": "male, young adult, high pitch",
+    "story_male": "male, young adult, moderate pitch",
+    "meme_male": "male, young adult, moderate pitch",
+    "calm_fun_male": "male, young adult, low pitch",
+}
+OMNIVOICE_EXTENDED = {"meme_male_gasp", "meme_male_normal", "meme_male_excited", "meme_male_confident",
+                       "fun_male_gasp", "story_male_normal", "calm_fun_male_sad", "calm_fun_male_whispering",
+                       "meme_male_sad", "meme_male_whispering", "fun_male_excited"}
+OMNIVOICE_ALL = set(OMNIVOICE_PRESETS.keys()) | OMNIVOICE_EXTENDED
+
+def _is_omnivoice_voice(v: str) -> bool:
+    v = str(v or "").strip()
+    if v in OMNIVOICE_ALL:
+        return True
+    for p in OMNIVOICE_PRESETS:
+        if v.startswith(p):
+            return True
+    return False
+
+def _resolve_omnivoice_preset_cloud(voice: str) -> str:
+    v = str(voice or "").strip()
+    if v in OMNIVOICE_PRESETS:
+        return v
+    for p in OMNIVOICE_PRESETS:
+        if v.startswith(p):
+            return p
+    return "meme_male"
+
 
 class CloudTTSEngine:
     def __init__(self, log_fn: Optional[Callable[[str], None]] = None, ffmpeg_path: str = "ffmpeg"):
         self.log_fn = log_fn or print
         self.ffmpeg_path = ffmpeg_path
         self.normalizer = VietnameseTextNormalizer()
+        # OmniVoice Cloud config — đọc từ config.json hoặc env
+        self.omnivoice_api_url = os.getenv("OMNIVOICE_API_URL", "https://hoangtaiii-omnivoice.hf.space").rstrip("/")
+        self.omnivoice_api_key = os.getenv("OMNIVOICE_API_KEY", "sk-demo123")
+        # Thử đọc config.json nếu có
+        try:
+            cfg_path = Path(__file__).resolve().parents[2] / "config.json"
+            if cfg_path.exists():
+                cfg = json.loads(cfg_path.read_text(encoding="utf-8"))
+                tts_cfg = cfg.get("tts", {})
+                self.omnivoice_api_url = (tts_cfg.get("omnivoice_api_url") or tts_cfg.get("omnivoice_cloud_url") or self.omnivoice_api_url).rstrip("/")
+                self.omnivoice_api_key = tts_cfg.get("omnivoice_api_key") or tts_cfg.get("omnivoice_cloud_api_key") or self.omnivoice_api_key
+        except Exception:
+            pass
 
     def _log(self, msg: str):
         self.log_fn(f"[Cloud TTS] {msg}")
@@ -57,16 +100,34 @@ class CloudTTSEngine:
             self._log("⚠️ Không có block phụ đề nào để tổng hợp TTS.")
             return False
 
-        self._log(f"⚡ Tạo giọng đọc song song cho {len(blocks)} câu thoại ({voice})...")
-
-        loop = asyncio.new_event_loop()
-        asyncio.set_event_loop(loop)
-        try:
-            success = loop.run_until_complete(
-                self._synthesize_blocks_parallel(blocks, segments_dir, voice, speed, pitch, volume)
-            )
-        finally:
-            loop.close()
+        # ── Auto-dispatch: OmniVoice Cloud nếu voice là preset mới ──
+        is_ov = _is_omnivoice_voice(voice)
+        if is_ov:
+            self._log(f"🎙️ Phát hiện giọng OmniVoice Cloud: {voice} -> {self.omnivoice_api_url}")
+            try:
+                import requests  # check sẵn
+            except ImportError:
+                self._log("❌ requests chưa cài — pip install requests")
+                return False
+            self._log(f"⚡ Tạo giọng đọc OmniVoice Cloud song song cho {len(blocks)} câu ({voice})...")
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            try:
+                success = loop.run_until_complete(
+                    self._synthesize_blocks_parallel_omnivoice(blocks, segments_dir, voice, speed)
+                )
+            finally:
+                loop.close()
+        else:
+            self._log(f"⚡ Tạo giọng đọc song song cho {len(blocks)} câu thoại ({voice})...")
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            try:
+                success = loop.run_until_complete(
+                    self._synthesize_blocks_parallel(blocks, segments_dir, voice, speed, pitch, volume)
+                )
+            finally:
+                loop.close()
 
         if not success:
             self._log("❌ Lỗi tổng hợp giọng đọc TTS.")
@@ -140,6 +201,80 @@ class CloudTTSEngine:
         await asyncio.gather(*tasks)
         return True
 
+    async def _synthesize_blocks_parallel_omnivoice(
+        self,
+        blocks: List[Dict],
+        segments_dir: Path,
+        voice: str,
+        speed: float = 1.0,
+    ) -> bool:
+        """
+        Gọi OmniVoice Cloud API song song — mỗi block 1 POST /v1/generate
+        Preset tự resolve từ voice (fun_male, meme_male_gasp,...) -> base preset
+        Nếu API lỗi 503/429 sẽ auto-fallback sang Edge-TTS sau 3 thử
+        """
+        import requests
+        preset = _resolve_omnivoice_preset_cloud(voice)
+        # speed mapping: clamp 0.75-1.65
+        speed_val = max(0.75, min(1.65, float(speed) if speed else 1.08))
+        semaphore = asyncio.Semaphore(4)  # HF ZeroGPU giới hạn, không spam 8
+
+        def _do_request_sync(text: str, wav_path: Path, seg_path: Path = None):
+            url = f"{self.omnivoice_api_url}/v1/generate"
+            headers = {"X-API-Key": self.omnivoice_api_key, "Content-Type": "application/json"}
+            payload = {"text": text, "preset": preset, "speed": speed_val, "steps": 16, "guidance": 2.0, "language": "vi"}
+            resp = requests.post(url, headers=headers, json=payload, timeout=120)
+            if resp.status_code != 200:
+                raise RuntimeError(f"Cloud {resp.status_code}: {resp.text[:300]}")
+            # Lưu wav tạm rồi convert chuẩn 16k mono
+            tmp_wav = segments_dir / f"tmp_{wav_path.stem}.wav"
+            tmp_wav.write_bytes(resp.content)
+            cmd = [str(self.ffmpeg_path), "-y", "-i", str(tmp_wav), "-ar", "16000", "-ac", "1", str(wav_path)]
+            subprocess.run(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True)
+            if tmp_wav.exists():
+                try: tmp_wav.unlink()
+                except: pass
+            if seg_path and seg_path.exists():
+                try: seg_path.unlink()
+                except: pass
+
+        async def _synthesize_one(block: Dict):
+            raw_text = block["text"].strip()
+            if not raw_text:
+                return
+            text = self.normalizer.normalize(raw_text) if hasattr(self.normalizer, "normalize") else raw_text
+            text = self._sanitize_for_tts(text)
+            wav_path = segments_dir / f"seg_{block['id']:04d}.wav"
+            if wav_path.exists() and wav_path.stat().st_size > 500:
+                return
+            async with semaphore:
+                for attempt in range(3):
+                    try:
+                        loop = asyncio.get_event_loop()
+                        await loop.run_in_executor(None, _do_request_sync, text, wav_path, None)
+                        break
+                    except Exception as e:
+                        if attempt == 2:
+                            self._log(f"⚠️ OmniVoice Cloud block {block['id']} thất bại: {e} -> fallback Edge-TTS")
+                            # Fallback sang Edge-TTS cho block này
+                            try:
+                                import edge_tts
+                                rate_str = f"{int(round((speed_val - 1.0) * 100)):+d}%"
+                                seg_mp3 = segments_dir / f"seg_{block['id']:04d}.mp3"
+                                communicate = edge_tts.Communicate(text, voice="vi-VN-NamMinhNeural", rate=rate_str)
+                                await communicate.save(str(seg_mp3))
+                                cmd2 = [str(self.ffmpeg_path), "-y", "-i", str(seg_mp3), "-ar", "16000", "-ac", "1", str(wav_path)]
+                                subprocess.run(cmd2, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True)
+                                if seg_mp3.exists(): seg_mp3.unlink()
+                            except Exception as e2:
+                                self._log(f"⚠️ Fallback Edge-TTS cũng lỗi block {block['id']}: {e2}")
+                        else:
+                            await asyncio.sleep(1.0 * (attempt + 1))
+
+        tasks = [_synthesize_one(b) for b in blocks]
+        await asyncio.gather(*tasks)
+        return True
+
     def _merge_segments_to_timeline(self, blocks: List[Dict], segments_dir: Path, output_wav: Path) -> bool:
         try:
             from pydub import AudioSegment

app/core/tts_worker_cli.py

@@ -234,6 +234,51 @@ def download_piper_model(voice_name, dest_dir):
     urllib.request.urlretrieve(f"{base_url}.onnx", str(onnx_file))
     return onnx_file, json_file
 
+# ── OmniVoice Presets — 100% đồng bộ với D:\omnivoice - web\app.py ──
+OMNIVOICE_PRESETS = {
+    "fun_male": "male, young adult, high pitch",
+    "story_male": "male, young adult, moderate pitch",
+    "meme_male": "male, young adult, moderate pitch",
+    "calm_fun_male": "male, young adult, low pitch",
+}
+# Extended mapping: voice_name -> (base_preset, emotion_tag)
+OMNIVOICE_EXTENDED = {
+    "meme_male_gasp": ("meme_male", "gasp"),
+    "meme_male_normal": ("meme_male", "normal"),
+    "meme_male_excited": ("meme_male", "excited"),
+    "meme_male_confident": ("meme_male", "confident"),
+    "meme_male_sad": ("meme_male", "sad"),
+    "meme_male_whispering": ("meme_male", "whispering"),
+    "fun_male_gasp": ("fun_male", "gasp"),
+    "fun_male_excited": ("fun_male", "excited"),
+    "story_male_normal": ("story_male", "normal"),
+    "calm_fun_male_sad": ("calm_fun_male", "sad"),
+    "calm_fun_male_whispering": ("calm_fun_male", "whispering"),
+}
+
+def _resolve_omnivoice_preset(voice: str, style: str = ""):
+    """
+    Resolve voice string -> (preset_key, instruct).
+    Ưu tiên: voice trực tiếp là preset -> dùng luôn.
+    Nếu voice là extended (meme_male_gasp...) -> lấy base preset.
+    Nếu style override -> dùng style để chọn tag.
+    Returns (preset_key, instruct, is_extended)
+    """
+    v = str(voice or "").strip()
+    s = str(style or "").strip().lower()
+    # direct preset
+    if v in OMNIVOICE_PRESETS:
+        return v, OMNIVOICE_PRESETS[v], False
+    if v in OMNIVOICE_EXTENDED:
+        base, _ = OMNIVOICE_EXTENDED[v]
+        return base, OMNIVOICE_PRESETS.get(base, "male, young adult, moderate pitch"), True
+    # fallback: try prefix match
+    for preset in OMNIVOICE_PRESETS:
+        if v.startswith(preset):
+            return preset, OMNIVOICE_PRESETS[preset], True
+    # default
+    return "meme_male", OMNIVOICE_PRESETS["meme_male"], False
+
 def _omnivoice_tag_for_text(text, style="", voice=""):
     """
     Determine OmniVoice emotion tag with priority:
@@ -292,6 +337,9 @@ def _omnivoice_tag_for_text(text, style="", voice=""):
     return tag, clean_text or text
 
 def _omnivoice_params(tag, tts_config):
+    # Base presets đồng bộ với omnivoice-web + emotion overrides
+    # Khi voice là fun_male/story_male/... ta sẽ ưu tiên instruct của preset đó
+    # trước rồi mới apply tag-based speed/guidance tweak
     preset = {
         "normal": ("male, young adult, moderate pitch", 1.06, 20, 2.0),
         "gasp": ("male, young adult, high pitch", 1.15, 20, 2.2),
@@ -302,7 +350,22 @@ def _omnivoice_params(tag, tts_config):
         "confident": ("male, young adult, low pitch", 1.02, 20, 2.0),
         "playful": ("male, young adult, high pitch", 1.08, 20, 2.1),
     }
+    # Nếu tts_config có omnivoice_preset (fun_male...) ưu tiên instruct preset đó
+    base_preset_key = tts_config.get("omnivoice_preset", "")
+    base_instruct = OMNIVOICE_PRESETS.get(base_preset_key)
     instruct, speed, steps, guidance = preset.get(tag, preset["normal"])
+    # Override instruct nếu có base preset
+    if base_instruct:
+        instruct = base_instruct
+        # tweak speed/guidance theo tag nhưng giữ instruct của preset
+        if tag == "gasp":
+            speed, guidance = 1.15, 2.2
+        elif tag == "excited":
+            speed, guidance = 1.16, 2.3
+        elif tag == "sad":
+            speed, guidance = 0.90, 1.8
+        elif tag == "whispering":
+            speed, guidance = 0.95, 1.6
     return {
         "instruct": tts_config.get("omnivoice_instruct", instruct),
         "speed": float(tts_config.get("omnivoice_speed", speed)),
@@ -496,12 +559,27 @@ except Exception as e:
                 pass
 
 def _omnivoice_request_params(text, args, tts_config):
+    # Resolve preset từ voice — nếu voice là preset mới (fun_male...) thì inject vào tts_config để _omnivoice_params ưu tiên
+    preset_key, preset_instruct, is_extended = _resolve_omnivoice_preset(args.voice, args.style)
+    # Clone tts_config để không mutate global
+    cfg = dict(tts_config)
+    cfg["omnivoice_preset"] = preset_key
+    # Nếu tts_config chưa có instruct và preset_instruct khác default thì ưu tiên preset instruct
+    if not cfg.get("omnivoice_instruct"):
+        # chỉ set nếu tag-based instruct chưa bị override bởi user
+        pass
     tag, clean_text = _omnivoice_tag_for_text(text, style=args.style, voice=args.voice)
-    params = _omnivoice_params(tag, tts_config)
+    # Nếu voice là preset thuần túy và style == auto/default thì override tag instruct bằng preset instruct
+    # để giữ đúng chất giọng fun_male/story_male...
+    base_tag_params = _omnivoice_params(tag, cfg)
+    # Nếu voice là preset trực tiếp (không extended) và tag == normal, ép instruct = preset_instruct
+    if not is_extended and str(args.style or "").lower() in ("", "auto", "default") and tag == "normal":
+        base_tag_params["instruct"] = preset_instruct
+    params = base_tag_params
     try:
         speed_multiplier = float(args.speed)
         if speed_multiplier > 0:
-            params["speed"] = max(0.75, min(1.35, params["speed"] * speed_multiplier))
+            params["speed"] = max(0.75, min(1.65, params["speed"] * speed_multiplier))
     except Exception:
         pass
     return tag, clean_text, params
@@ -598,8 +676,59 @@ except Exception as e:
         except Exception:
             pass
 
+def run_omnivoice_cloud_tts(text, output_path, args, tts_config):
+    """
+    Goi OmniVoice Cloud API (HF Space https://hoangtaiii-omnivoice.hf.space).
+    Dong bo 100% voi omnivoice-web PRESETS (fun_male, story_male, meme_male, calm_fun_male).
+    Tham khao HUONG_DAN_KET_NOI_API.md
+    """
+    import requests
+    api_url = tts_config.get("omnivoice_api_url", "") or tts_config.get("omnivoice_cloud_url", "") or os.getenv("OMNIVOICE_API_URL", "https://hoangtaiii-omnivoice.hf.space")
+    api_key = tts_config.get("omnivoice_api_key", "") or tts_config.get("omnivoice_cloud_api_key", "") or os.getenv("OMNIVOICE_API_KEY", "sk-demo123")
+    api_url = str(api_url).rstrip("/")
+    # Resolve preset từ voice
+    preset_key, _, _ = _resolve_omnivoice_preset(args.voice, args.style)
+    # Nếu voice là extended (meme_male_gasp) thì gửi preset = base preset
+    # Cloud sẽ tự dùng default voice nếu preset == default và không truyền instruct
+    payload = {
+        "text": str(text).strip(),
+        "preset": preset_key if preset_key in OMNIVOICE_PRESETS else "meme_male",
+        "speed": float(args.speed) if str(args.speed).replace('.','',1).isdigit() else 1.08,
+        "steps": 16,
+        "guidance": 2.0,
+        "language": "vi"
+    }
+    # Nếu tts_config có instruct override
+    if tts_config.get("omnivoice_instruct"):
+        payload["instruct"] = tts_config["omnivoice_instruct"]
+    headers = {"X-API-Key": api_key, "Content-Type": "application/json"}
+    output_path = Path(output_path)
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    print(f"[OMNIVOICE CLOUD] POST {api_url}/v1/generate preset={payload['preset']} text_len={len(payload['text'])}")
+    resp = requests.post(f"{api_url}/v1/generate", headers=headers, json=payload, timeout=int(tts_config.get("omnivoice_timeout_seconds", 240)))
+    if resp.status_code != 200:
+        raise RuntimeError(f"OmniVoice Cloud {resp.status_code}: {resp.text[:500]}")
+    ctype = resp.headers.get("content-type", "")
+    if "audio" not in ctype and len(resp.content) < 1000:
+        # có thể là JSON lỗi
+        try:
+            j = resp.json()
+            raise RuntimeError(f"OmniVoice Cloud JSON error: {j}")
+        except Exception:
+            pass
+    output_path.write_bytes(resp.content)
+    # verify
+    if not verify_audio_file(output_path):
+        raise RuntimeError("OmniVoice Cloud returned invalid audio (<500 bytes or silence)")
+    print(f"[OMNIVOICE CLOUD] Saved {output_path} ({output_path.stat().st_size} bytes)")
+
 def synthesize_text_to_wav(text, output_path, args, tts_config, piper_voice=None, voice_override=None, omnivoice_session=None):
-    engine = args.engine.lower()
+    engine = str(args.engine or "").lower().strip()
+    # Chuẩn hóa alias: "omnivoice (local)" -> "omnivoice", "omnivoice cloud (hf space)" -> "omnivoice_cloud"
+    if "cloud" in engine or "hf" in engine or "api" in engine:
+        engine = "omnivoice_cloud"
+    elif "omnivoice" in engine:
+        engine = "omnivoice"
     voice = voice_override or args.voice
     if engine == "piper" and piper_voice:
         import wave
@@ -614,6 +743,15 @@ def synthesize_text_to_wav(text, output_path, args, tts_config, piper_voice=None
             piper_voice.synthesize_wav(text, wav_file, syn_config=syn_config)
     elif engine == "omnivoice":
         run_omnivoice_tts(text, output_path, args, tts_config, omnivoice_session=omnivoice_session)
+    elif engine == "omnivoice_cloud":
+        # Nếu args.voice bị override bởi fallback, tạm gán lại args.voice = voice để resolve preset đúng
+        orig_voice = args.voice
+        try:
+            if voice_override:
+                args.voice = voice_override
+            run_omnivoice_cloud_tts(text, output_path, args, tts_config)
+        finally:
+            args.voice = orig_voice
     else:
         asyncio.run(run_edge_tts(text, voice, output_path, args.speed, args.pitch, args.volume))
 
@@ -1070,7 +1208,10 @@ def main():
                 sys.exit(3)
 
     omnivoice_session = None
-    if args.engine.lower() == "omnivoice" and bool(tts_config.get("omnivoice_persistent_session", True)):
+    # Chỉ bật persistent session cho local OmniVoice; Cloud dùng HTTP không cần
+    _engine_norm = str(args.engine or "").lower()
+    _is_local_ov = ("omnivoice" in _engine_norm and "cloud" not in _engine_norm)
+    if _is_local_ov and bool(tts_config.get("omnivoice_persistent_session", True)):
         print("[OMNIVOICE SESSION] starting persistent model session...")
         omnivoice_session = OmniVoiceSession(args, tts_config, segments_dir)
         omnivoice_session.start()
@@ -1292,8 +1433,9 @@ def main():
                                 except Exception:
                                     pass
 
-        # --- Plan 4: Voice fallback ---
-        if not success and args.engine.lower() != "omnivoice" and tts_config.get("fallback_voice_enabled", True):
+        # --- Plan 4: Voice fallback (không áp dụng cho OmniVoice local/cloud — đã có retry riêng) ---
+        _is_ov = "omnivoice" in str(args.engine or "").lower()
+        if not success and not _is_ov and tts_config.get("fallback_voice_enabled", True):
             plans_tried.append("fallback_voice")
             fallback_voices = tts_config.get("fallback_voices", ["vi-VN-NamMinhNeural", "vi-VN-HoaiMyNeural", "vi-VN-HoaiAnNeural"])
             candidate_voices = [v for v in fallback_voices if v.lower() != args.voice.lower()]

app/main.py

@@ -566,7 +566,7 @@ class MainWindow(QMainWindow):
         engine = self.tts_engine_combo.currentText()
         self.tts_voice_combo.clear()
         if engine == "Edge-TTS":
-            self.tts_voice_combo.addItems(["vi-VN-HoaiMyNeural", "vi-VN-NamMinhNeural"])
+            self.tts_voice_combo.addItems(["vi-VN-HoaiMyNeural", "vi-VN-NamMinhNeural", "vi-VN-HoaiAnNeural"])
             self.tts_pitch_spin.setEnabled(True)
             if hasattr(self, "tts_style_combo"):
                 self.tts_style_combo.blockSignals(True)
@@ -574,21 +574,30 @@ class MainWindow(QMainWindow):
                 self.tts_style_combo.addItems(["Default"])
                 self.tts_style_combo.blockSignals(False)
                 self.tts_style_combo.setEnabled(False)
-        elif engine == "OmniVoice":
+        elif engine in ("OmniVoice", "OmniVoice (Local)", "OmniVoice Cloud (HF Space)"):
+            # Đồng bộ 100% với D:\omnivoice - web\app.py PRESETS + extended
             self.tts_voice_combo.addItems([
-                "meme_male_gasp",
-                "meme_male_normal",
-                "meme_male_excited",
-                "meme_male_confident",
+                # Preset gốc từ omnivoice-web (4 giọng chính)
+                "meme_male", "fun_male", "story_male", "calm_fun_male",
+                # Extended kết hợp style
+                "meme_male_gasp", "meme_male_normal", "meme_male_excited", "meme_male_confident",
+                "meme_male_sad", "meme_male_whispering",
+                "fun_male_gasp", "fun_male_excited",
+                "story_male_normal", "calm_fun_male_sad",
             ])
             self.tts_pitch_spin.setEnabled(False)
             if hasattr(self, "tts_style_combo"):
                 self.tts_style_combo.blockSignals(True)
                 self.tts_style_combo.clear()
                 # Default to "gasp" for OmniVoice (first item = default)
-                self.tts_style_combo.addItems(["gasp", "normal", "excited", "confident", "sad", "whispering", "auto"])
+                self.tts_style_combo.addItems(["auto", "gasp", "normal", "excited", "confident", "sad", "whispering", "sarcastic", "playful"])
                 self.tts_style_combo.blockSignals(False)
                 self.tts_style_combo.setEnabled(True)
+            # Tooltip cho Cloud
+            if engine == "OmniVoice Cloud (HF Space)":
+                self.tts_voice_combo.setToolTip("Gọi API https://hoangtaiii-omnivoice.hf.space — không cần GPU local, cần API key trong config.json (sk-demo123 mặc định)")
+            else:
+                self.tts_voice_combo.setToolTip("Chạy local qua OmniVoiceApp/.venv — cần GPU CUDA + model splendor1811/omnivoice-vietnamese")
         else: # Piper
             self.tts_voice_combo.addItems(["vi_VN-vais1000-medium", "vi_VN-vivos-x_low", "vi_VN-25hours_single-low"])
             self.tts_pitch_spin.setEnabled(False)
@@ -647,7 +656,7 @@ class MainWindow(QMainWindow):
             def run(self):
                 try:
                     from app.core.process_manager import ProcessManager
-                    timeout = 240 if engine == "omnivoice" else 30
+                    timeout = 240 if "omnivoice" in engine else 30
                     ProcessManager.instance().run_subprocess_sync(cmd, timeout=timeout)
                     self.finished_signal.emit(True)
                 except Exception as e:
@@ -686,9 +695,21 @@ class MainWindow(QMainWindow):
                         self.gpu_mode_combo.setCurrentIndex(idx_g)
 
                 engine = config.get("voice_engine", "Edge-TTS")
+                # Backward compat: OmniVoice -> OmniVoice (Local)
+                if engine == "OmniVoice":
+                    engine = "OmniVoice (Local)"
+                # tts.backend có thể là omnivoice_cloud -> map
+                if config.get("tts", {}).get("backend") == "omnivoice_cloud":
+                    engine = "OmniVoice Cloud (HF Space)"
                 idx = self.tts_engine_combo.findText(engine)
                 if idx != -1:
                     self.tts_engine_combo.setCurrentIndex(idx)
+                else:
+                    # fallback partial match
+                    for i in range(self.tts_engine_combo.count()):
+                        if engine.lower() in self.tts_engine_combo.itemText(i).lower() or self.tts_engine_combo.itemText(i).lower() in engine.lower():
+                            self.tts_engine_combo.setCurrentIndex(i)
+                            break
                     
                 self.on_tts_engine_changed()
                 
@@ -1680,7 +1701,7 @@ class MainWindow(QMainWindow):
 
         form = QVBoxLayout()
         self.tts_engine_combo = QComboBox()
-        self.tts_engine_combo.addItems(["Edge-TTS", "Piper", "OmniVoice"])
+        self.tts_engine_combo.addItems(["Edge-TTS", "OmniVoice (Local)", "OmniVoice Cloud (HF Space)", "Piper"])
         self.tts_engine_combo.currentTextChanged.connect(self.on_tts_engine_changed)
         self.tts_engine_combo.currentTextChanged.connect(self.save_voice_settings)
         form.addLayout(self._form_row("Voice engine", self.tts_engine_combo))
@@ -2282,7 +2303,7 @@ class MainWindow(QMainWindow):
         form.addRow("Translation Engine:", self.trans_combo)
         
         self.tts_engine_combo = QComboBox()
-        self.tts_engine_combo.addItems(["Edge-TTS", "Piper", "OmniVoice"])
+        self.tts_engine_combo.addItems(["Edge-TTS", "OmniVoice (Local)", "OmniVoice Cloud (HF Space)", "Piper"])
         self.tts_engine_combo.setSizePolicy(QSizePolicy.Policy.Expanding, QSizePolicy.Policy.Fixed)
         self.tts_engine_combo.setMinimumWidth(200)
         self.tts_engine_combo.currentTextChanged.connect(self.on_tts_engine_changed)

config.json

@@ -208,11 +208,21 @@
     "omnivoice_model": "splendor1811/omnivoice-vietnamese",
     "omnivoice_preset_vietnamese": "meme_male",
     "omnivoice_ref_audio": "C:\\Users\\Admin\\OmniVoiceApp\\outputs\\voice_clone_vi_exact.wav",
-    "omnivoice_ref_text": "\u00ca bro, chuy\u1ec7n n\u00e0y nghe vui thi\u1ec7t \u00e1! Tao v\u1eeba m\u00f2 ra c\u00e1ch l\u00e0m gi\u1ecdng n\u00e0y n\u00e8. Nghe th\u1eed \u0111i, \u0111\u1ea3m b\u1ea3o m\u00ea lu\u00f4n \u0111\u00f3, nghe ph\u00e1t cu\u1ed1n li\u1ec1n h\u00e0!",
+    "omnivoice_ref_text": "Ê bro, chuyện này nghe vui thiệt á! Tao vừa mò ra cách làm giọng này nè. Nghe thử đi, đảm bảo mê luôn đó, nghe phát cuốn liền hà!",
     "omnivoice_timeout_seconds": 240,
     "omnivoice_persistent_session": true,
     "omnivoice_session_ready_timeout_seconds": 300,
     "omnivoice_stage_timeout_seconds": 7200,
+    "omnivoice_api_url": "https://hoangtaiii-omnivoice.hf.space",
+    "omnivoice_api_key": "sk-demo123",
+    "omnivoice_cloud_url": "https://hoangtaiii-omnivoice.hf.space",
+    "omnivoice_cloud_api_key": "sk-demo123",
+    "omnivoice_presets": {
+      "fun_male": "male, young adult, high pitch",
+      "story_male": "male, young adult, moderate pitch",
+      "meme_male": "male, young adult, moderate pitch",
+      "calm_fun_male": "male, young adult, low pitch"
+    },
     "merge_short_segments_enabled": true,
     "merge_short_segments_min_input_blocks": 4,
     "merge_short_segments_max_gap_ms": 350,