# Default output options AUDIO_FORMAT = "m4b" AUDIO_BITRATE = "128k" LANGUAGE = "English" API_TIMEOUT = 600 # Timeout per chunk request in seconds MAX_RETRIES = 3 # Attempts per chunk request HEARTBEAT_INTERVAL_SECONDS = 30 # Print "still working" in console logs every N seconds # Words per TTS generation request. # Note that qwen-tts-demo does no chunking at all, but faster-qwen3-tts and # audio.cpp may do chunking as well, so you may be needlessly double-chunking. # This is the only size limit: there is no hard ceiling. However, servers # silently truncate audio when a single generation runs too long (roughly # ~2.5 min on the faster backend's static KV cache, ~11 min on the Gradio # demo) without reporting an error, so raising this is at your own risk. # The client-side truncation check still catches and retries gross cases. CHUNK_SIZE = 250 # Default TTS backend. # gradio: qwen-tts-demo # faster: faster-qwen-tts # audiocpp: audiocpp_server # The --backend CLI flag overrides this BACKEND = "gradio" ############################################################################### # BACKEND 1: qwen-tts-demo (gradio) options # ############################################################################### # There are different API URLs for CustomVoice and Base models so you can run both at once QWEN_API_URL = "http://127.0.0.1:7860" # CustomVoice model CLONE_API_URL = "http://127.0.0.1:7861" # Base model # Custom voice options SPEAKER = "Vivian" #Vivian, Serena, Uncle_Fu, Dylan, Eric, Ryan, Aiden, Ono_Anna, Sohee INSTRUCT = "Speak naturally and clearly, as if reading a dramatic book to an adult audience." # Don't clone with transcription, only use x-vector-only cloning. Generally "worse" XVECTOR_ONLY = False # Randomization seed. -1 means randomize with every generation # With SEED = -1 and CONSTANT_SEED = True, one random seed will be used for the entire audiobook. # This may keep the voice slightly more consistent across chunk boundaries SEED = -1 CONSTANT_SEED = False ############################################################################### # BACKEND 2: faster-qwen-tts options # ############################################################################### FASTER_API_URL = "http://127.0.0.1:8000" # faster-qwen3-tts server (Base model only) # Default voice if no --voice is passed FASTER_VOICE = "default" ############################################################################### # BACKEND 3: audio.cpp options # ############################################################################### AUDIOCPP_API_URL = "http://127.0.0.1:8080" # audio.cpp audiocpp_server # Model ids in the audio.cpp server.json config. AUDIOCPP_MODEL_ID = "qwen" AUDIOCPP_CLONE_MODEL_ID = "qwen-clone"