Grok Bot
76 lines
2.8 KiB
YAML
76 lines
2.8 KiB
YAML
# One settings file for the whole pipeline. Relative paths are relative to the repo root.
|
|
|
|
books_dir: books # where book folders live: books/<first-author>/<book>/
|
|
work_dir: /tmp/audiobook-work # scratch only (OCR output, wav pieces). Point at local NVMe in production.
|
|
|
|
# Where finished audio goes. Empty = the book folder in the repo (the agreed default).
|
|
# Set to a folder to write <archive_dir>/<first-author>/<book>/audio there instead.
|
|
archive_dir: ""
|
|
|
|
narrator: "" # spoken in the credits only if filled in (leave empty for a TTS voice)
|
|
|
|
# Inserted parts. All OFF by default. Per-book and per-chapter switches in book.json override these.
|
|
summaries:
|
|
chapter_preview: false
|
|
chapter_summary: false
|
|
concept_summary: false # one mid-chapter concept summary
|
|
|
|
# OpenAI-compatible chat endpoint, used only when a summary switch is on.
|
|
llm:
|
|
endpoint: "" # e.g. http://host:8081/v1/chat/completions
|
|
model: ""
|
|
api_key_env: "" # name of an environment variable that holds the key (never put a key here)
|
|
timeout_s: 300
|
|
|
|
# OpenAI-compatible speech endpoint.
|
|
voice:
|
|
endpoint: http://127.0.0.1:8000/v1/audio/speech
|
|
model: chatterbox
|
|
voice: default
|
|
extra: {} # extra JSON fields sent with every request (for example speed)
|
|
timeout_s: 600
|
|
max_chunk_chars: 300 # long paragraphs are cut at sentence ends into pieces this size or smaller
|
|
gap_sentence_s: 0.25 # pause between pieces inside a paragraph
|
|
gap_paragraph_s: 0.7 # pause between paragraphs and after cues and labels
|
|
|
|
ocr:
|
|
enabled: true # run ocrmypdf only when a page has no text layer
|
|
language: eng
|
|
min_chars_per_page: 10 # fewer characters than this = no text layer
|
|
|
|
script:
|
|
retail_sample_words: 400 # about 2.5 minutes at 155 words per minute (ACX asks for 1 to 5 minutes)
|
|
chime_label: Chime
|
|
|
|
license:
|
|
public_domain_max_year: 1930 # US, as of 2026
|
|
|
|
audio:
|
|
sample_rate: 44100
|
|
channels: 1
|
|
mp3_bitrate: 192k
|
|
head_room_tone_s: 0.75 # ACX 0.5 to 1
|
|
tail_room_tone_s: 2.0 # ACX 1 to 5
|
|
room_tone_db: -66 # RMS level of the room tone (ACX noise floor limit is -60)
|
|
target_rms_db: -20.5 # ACX -23 to -18
|
|
peak_ceiling_db: -4.0 # limiter ceiling; ACX limit is -3 and MP3 encoding can add a little
|
|
chime_hz: 660
|
|
chime_s: 1.0
|
|
chime_db: -18 # peak level of the chime in the final file, kept soft
|
|
|
|
acx: # limits used by the checker (R13)
|
|
min_bitrate_kbps: 192
|
|
sample_rate: 44100
|
|
rms_min_db: -23
|
|
rms_max_db: -18
|
|
peak_max_db: -3
|
|
noise_floor_max_db: -60
|
|
head_min_s: 0.5
|
|
head_max_s: 1.0
|
|
tail_min_s: 1.0
|
|
tail_max_s: 5.0
|
|
max_minutes: 120
|
|
retail_min_s: 60
|
|
retail_max_s: 300
|
|
speech_window_db: -50 # a 50 ms window louder than this counts as speech when measuring head and tail
|