fix: reduce chunk size to 40

1 files changed, 11 insertions, 4 deletions

diff --git a/converter/config.py b/converter/config.py
index efec0dd..dcec145 100644
--- a/converter/config.py
+++ b/converter/config.py
@@ -123,8 +123,9 @@ FASTER_TTS_API_URL = "http://127.0.0.1:8000"
 # name is unknown, so a mismatch here is easy to miss.
 FASTER_TTS_VOICE = "default"
 FASTER_TTS_SAMPLE_RATE = 24000  # Qwen3-TTS 12Hz codec output rate
-# The Gradio demo sub-chunked text server-side (~200 chars); the faster server
-# takes one generation per request, so long chunks are sub-chunked client-side.
+# Safety net: the faster server takes one generation per request, so any
+# chunk longer than this is sub-chunked client-side (a no-op at the default
+# CHUNK_SIZE_WORDS above; kept in case the chunk size is ever raised).
 FASTER_SUBCHUNK_WORDS = 40  # ~200 chars per request
 FASTER_HTTP_TIMEOUT = 300  # Seconds before a speech request times out
 FASTER_SUBCHUNK_RETRIES = 3  # Attempts per sub-chunk request
@@ -138,8 +139,14 @@ AUDIOBOOKS_FOLDER = BASE_DIR / "output"  # Output folder
 CHUNKS_FOLDER = BASE_DIR / "chunks"  # Scratch space for per-chunk audio (cleaned per book)
 LOGS_FOLDER = BASE_DIR / "logs"
 
-CHUNK_SIZE_WORDS = 1500  # Words per TTS chunk
-MIN_DELAY_BETWEEN_CHUNKS = 1  # Seconds between API calls
+# Words per TTS generation request. Each API call is ONE model generation:
+# long generations lose prosody, can degrade into garbled audio, and text
+# past the model's token limit is never spoken. ~40 words (~200 chars) is
+# the per-request length the old qwen-tts demo enforced server-side.
+CHUNK_SIZE_WORDS = 40
+# Pause between API calls (rate-limit protection for hosted demos; a local
+# server needs no delay).
+MIN_DELAY_BETWEEN_CHUNKS = 0
 HEARTBEAT_INTERVAL_SECONDS = 30  # Print "still working" this often during a chunk
 
 # =============================================================================