fix: do not chunk with audio.cpp backend (double chunking)

1 files changed, 36 insertions, 4 deletions

diff --git a/converter/converter.py b/converter/converter.py
index 6994a31..f1064f1 100644
--- a/converter/converter.py
+++ b/converter/converter.py
@@ -134,7 +134,8 @@ class AudiobookConverter:
                  voice_clone_ref_text: Optional[str] = None, skip_transcription: bool = False,
                  speed: float = 1.0, single_file: bool = False, output_format: str = config.AUDIO_FORMAT,
                  language: Optional[str] = None, backend: str = BACKEND_GRADIO,
-                 voice: Optional[str] = None, debug: bool = False):
+                 voice: Optional[str] = None, debug: bool = False,
+                 chunk: bool = False):
         if speed <= 0:
             raise ValueError(f"Speed must be a positive number, got {speed}")
         if output_format not in AUDIO_FORMATS:
@@ -154,6 +155,12 @@ class AudiobookConverter:
         self.backend = backend
         self.voice = voice
         self.debug = bool(debug)
+        # Client-side chunking: the gradio and faster backends always chunk
+        # (their servers do one generation per request and silently truncate
+        # long text). The audio.cpp server chunks long text itself, so it
+        # defaults to one request per chapter; --chunk forces client-side
+        # chunking on top (possible needless double-chunking).
+        self.client_chunks = bool(chunk) or backend != BACKEND_AUDIOCPP
         self._validate_configuration()
         if backend == BACKEND_FASTER:
             # The faster backend always voice-clones using a reference voice
@@ -162,7 +169,8 @@ class AudiobookConverter:
         elif backend == BACKEND_AUDIOCPP:
             # Speaker mode (no voice) uses a built-in CustomVoice speaker;
             # an explicit voice selects a server-side preset (cloning).
-            self.tts = AudioCppTTSClient(voice=voice, language=self.language)
+            self.tts = AudioCppTTSClient(voice=voice, language=self.language,
+                                         chunk_text=self.client_chunks)
         else:
             self.tts = QwenTTSClient(
                 voice_mode=voice_mode,
@@ -426,6 +434,18 @@ class AudiobookConverter:
         logger.info("Chunk processing completed: %d/%d chunks", successful_chunks, total_chunks)
         return results
 
+    def _chapter_chunks(self, text: str) -> List[str]:
+        """Split chapter text into TTS requests.
+
+        Client-side chunking splits into CHUNK_SIZE-word chunks (gradio and
+        faster always; audio.cpp only with --chunk). Otherwise (audio.cpp
+        default) the whole text is one request and the server does its own
+        long-form chunking.
+        """
+        if self.client_chunks:
+            return chunking.split_into_chunks(text)
+        return [text] if text.strip() else []
+
     def _convert_text(self, text: str, output_path: Path, start_time: float,
                       speed: Optional[float] = None,
                       output_format: Optional[str] = None,
@@ -452,7 +472,7 @@ class AudiobookConverter:
 
             logger.info("Extracted %d characters (%d words)", len(text), len(text.split()))
 
-            chunks = chunking.split_into_chunks(text)
+            chunks = self._chapter_chunks(text)
             total_chunks = len(chunks)
             if total_chunks == 0:
                 logger.error("No chunks created")
@@ -460,7 +480,13 @@ class AudiobookConverter:
 
             chunk_sizes = [len(chunk.split()) for chunk in chunks]
             avg_chunk_size = sum(chunk_sizes) / len(chunk_sizes)
-            logger.info("Split into %d chunks (avg %.0f words per chunk)", total_chunks, avg_chunk_size)
+            if len(chunks) == 1:
+                logger.info("Sending the whole text as one request (%d words; "
+                            "the server chunks long text itself)",
+                            chunk_sizes[0])
+            else:
+                logger.info("Split into %d chunks (avg %.0f words per chunk)",
+                            total_chunks, avg_chunk_size)
             backend_labels = {
                 BACKEND_FASTER: "faster TTS API",
                 BACKEND_AUDIOCPP: "audio.cpp server",
@@ -526,6 +552,12 @@ class AudiobookConverter:
             else:
                 print("Backend: audio.cpp (custom voice, built-in speaker)")
                 print(f"Speaker: {config.SPEAKER}")
+            if self.client_chunks:
+                print("Chunking: client-side (--chunk; the server also chunks "
+                      "long text itself, so this may double-chunk)")
+            else:
+                print("Chunking: server-side (one request per chapter; "
+                      "--chunk forces client-side chunking)")
             print(f"Language: {self.language}")
         else:
             api_url = (config.CLONE_API_URL if self.voice_mode == VOICE_MODE_CLONE