unify voice caching: VoiceCache class used by both WebUI and PyQt

- domain/voice_loader.py: VoiceCache class now used by both UIs;
  resolve_voice() and load_voice_cached() accept VoiceCache or plain dict;
  added hasattr(pipeline, 'load_single_voice') safety check from WebUI
- conversion_runner.py: replaced local _resolve_voice() with domain's
  resolve_voice(); voice_cache changed from Dict to VoiceCache instance;
  all cache access uses VoiceCache.get()/set() API
- pyqt/conversion.py: self.voice_cache changed from Dict to VoiceCache
- debug_tts_runner.py: imports resolve_voice from domain instead of
  removed _resolve_voice from conversion_runner

abogen/domain/voice_loader.py

@@ -65,7 +65,7 @@ def resolve_voice(
 
     # Load voice
     if "*" in voice_spec:
-        if pipeline is None:
+        if pipeline is None or not hasattr(pipeline, "load_single_voice"):
             return voice_spec
         loaded_voice = get_new_voice(pipeline, voice_spec, use_gpu)
     else:
@@ -82,7 +82,7 @@ def load_voice_cached(
     voice_name: str,
     pipeline: Any,
     use_gpu: bool,
-    cache: Optional[Dict[str, Any]] = None,
+    cache: Any = None,
 ) -> Any:
     """Load voice with caching (compatibility wrapper for PyQt).
 
@@ -93,24 +93,32 @@ def load_voice_cached(
         voice_name: Voice name or formula string.
         pipeline: TTS pipeline instance.
         use_gpu: Whether to use GPU.
-        cache: Optional dict to use as cache (instead of VoiceCache).
+        cache: Optional VoiceCache or dict to use as cache.
 
     Returns:
         Loaded voice tensor or voice name string.
     """
-    # Use dict cache if provided (for backward compatibility)
+    # Check cache (supports both VoiceCache and plain dict)
     if cache is not None:
-        if voice_name in cache:
+        if isinstance(cache, VoiceCache):
+            if cache.contains(voice_name):
+                return cache.get(voice_name)
+        elif voice_name in cache:
             return cache[voice_name]
 
     # Load voice
     if "*" in voice_name:
+        if pipeline is None or not hasattr(pipeline, "load_single_voice"):
+            return voice_name
         loaded_voice = get_new_voice(pipeline, voice_name, use_gpu)
     else:
         loaded_voice = voice_name
 
     # Cache it
     if cache is not None:
+        if isinstance(cache, VoiceCache):
+            cache.set(voice_name, loaded_voice)
+        else:
             cache[voice_name] = loaded_voice
 
     return loaded_voice

abogen/pyqt/conversion.py

@@ -44,7 +44,7 @@ from abogen.domain.audio_buffer import (
     SAMPLE_RATE,
 )
 from abogen.domain.subtitle_generation import process_subtitle_tokens
-from abogen.domain.voice_loader import load_voice_cached, resolve_voice
+from abogen.domain.voice_loader import VoiceCache, load_voice_cached, resolve_voice
 from abogen.domain.progress import calc_etr_str
 from abogen.domain.normalization import prepare_text_for_tts
 from abogen.domain.pronunciation import (
@@ -295,7 +295,7 @@ class ConversionThread(QThread):
         self.use_spacy_segmentation = True  # Default, will be overridden from GUI
         # Set split pattern based on language and subtitle mode
         self.split_pattern = get_split_pattern(lang_code, subtitle_mode)
-        self.voice_cache = {}  # Cache for loaded voices
+        self.voice_cache = VoiceCache()  # Cache for loaded voices
 
     def load_voice_cached(self, voice_name, tts):
         """Load voice with caching to avoid reloading same voice.

abogen/webui/conversion_runner.py

@@ -119,6 +119,7 @@ from abogen.domain.audio_buffer import (
 from abogen.domain.audio_sink import AudioSink, open_audio_sink
 from abogen.domain.pipeline_factory import PipelinePool
 from abogen.domain.conversion_engine import run_tts_segment_loop, process_and_write_subtitles, SegmentStats
+from abogen.domain.voice_loader import VoiceCache, resolve_voice
 from abogen.domain.voice_utils import resolve_voice_target as _resolve_voice_target
 
 
@@ -216,11 +217,11 @@ def run_conversion_job(job: Job) -> None:
 
             if provider == "kokoro":
                 kokoro_backend = pipeline_pool.get("kokoro", job.language, job.use_gpu, job=job)
-                choice = _resolve_voice(kokoro_backend, resolved, job.use_gpu)
+                choice = resolve_voice(resolved, kokoro_backend, job.use_gpu, cache=voice_cache)
             else:
                 choice = resolved
 
-            voice_cache[cache_key] = choice
+            voice_cache.set(cache_key, choice)
             return provider, resolved, choice, speed, steps
 
         extraction = extract_from_path(job.stored_path)
@@ -360,7 +361,7 @@ def run_conversion_job(job: Job) -> None:
             chapter_dir.mkdir(parents=True, exist_ok=True)
 
         base_voice_spec = _job_voice_fallback(job)
-        voice_cache: Dict[str, Any] = {}
+        voice_cache = VoiceCache()
         base_provider, base_voice_resolved, _, _ = _resolve_voice_target(
             base_voice_spec, normalized_profiles,
             job_voice=getattr(job, "voice", "M1"),
@@ -368,7 +369,7 @@ def run_conversion_job(job: Job) -> None:
         )
         if base_provider == "kokoro" and base_voice_resolved and "*" not in base_voice_resolved:
             kokoro_backend = pipeline_pool.get("kokoro", job.language, job.use_gpu, job=job)
-            voice_cache[f"kokoro:{base_voice_resolved}"] = _resolve_voice(kokoro_backend, base_voice_resolved, job.use_gpu)
+            voice_cache.set(f"kokoro:{base_voice_resolved}", resolve_voice(base_voice_resolved, kokoro_backend, job.use_gpu))
         processed_chars = 0
         current_time = 0.0
         etr_start_time = time.time()
@@ -550,8 +551,7 @@ def run_conversion_job(job: Job) -> None:
                 voice_choice = voice_cache.get(chapter_cache_key)
                 if voice_choice is None:
                     kokoro_backend = pipeline_pool.get("kokoro", job.language, job.use_gpu, job=job)
-                    voice_choice = _resolve_voice(kokoro_backend, chapter_voice_resolved, job.use_gpu)
-                    voice_cache[chapter_cache_key] = voice_choice
+                    voice_choice = resolve_voice(chapter_voice_resolved, kokoro_backend, job.use_gpu, cache=voice_cache)
             else:
                 voice_choice = chapter_voice_resolved
 
@@ -694,12 +694,12 @@ def run_conversion_job(job: Job) -> None:
                             chunk_voice_choice = voice_cache.get(chunk_cache_key)
                             if chunk_voice_choice is None:
                                 kokoro_backend = pipeline_pool.get("kokoro", job.language, job.use_gpu, job=job)
-                                chunk_voice_choice = _resolve_voice(
-                                    kokoro_backend,
+                                chunk_voice_choice = resolve_voice(
                                     chunk_voice_resolved,
+                                    kokoro_backend,
                                     job.use_gpu,
+                                    cache=voice_cache,
                                 )
-                                voice_cache[chunk_cache_key] = chunk_voice_choice
                         else:
                             chunk_voice_choice = chunk_voice_resolved
 
@@ -1066,14 +1066,6 @@ def _prepare_project_layout(job: Job, base_dir: Path) -> tuple[Path, Path, Path,
 
 
 
-def _resolve_voice(pipeline, voice_spec: str, use_gpu: bool):
-    if "*" in voice_spec:
-        if pipeline is None or not hasattr(pipeline, "load_single_voice"):
-            return voice_spec
-        return get_new_voice(pipeline, voice_spec, use_gpu)
-    return voice_spec
-
-
 
 def _make_canceller(job: Job) -> Callable[[], None]:
     def _cancel() -> None:

abogen/webui/debug_tts_runner.py

@@ -14,7 +14,8 @@ from abogen.kokoro_text_normalization import normalize_for_pipeline
 from abogen.normalization_settings import build_apostrophe_config
 from abogen.text_extractor import extract_from_path
 from abogen.voice_cache import ensure_voice_assets
-from abogen.webui.conversion_runner import SAMPLE_RATE, _select_device, _to_float32, _resolve_voice, _spec_to_voice_ids
+from abogen.webui.conversion_runner import SAMPLE_RATE, _select_device, _to_float32, _spec_to_voice_ids
+from abogen.domain.voice_loader import resolve_voice
 from abogen.domain.split_pattern import get_split_pattern
 from abogen.tts_plugin.utils import create_pipeline
 
@@ -176,7 +177,7 @@ def run_debug_tts_wavs(
             pass
 
     pipeline = _load_pipeline(language, use_gpu)
-    voice_choice = _resolve_voice(pipeline, voice_spec, use_gpu)
+    voice_choice = resolve_voice(voice_spec, pipeline, use_gpu)
 
     apostrophe_config = build_apostrophe_config(settings=settings)
     normalization_settings = dict(settings)