Check if a part file exists and return its duration, or None.

Read nested key from user_prefs.json. Returns None if missing/unreadable.

Resolve voice with precedence: env var > user_prefs.json > hardcoded default.

Save prefs to disk, stamping updated_at with UTC ISO timestamp.

Load user_prefs.template.json as default structure.

Add or update an entry in prefs["design_references"].

Split text into chunks for TTS synthesis. Handles both Chinese (。；) and English (. ; ? !) sentence boundaries. Sentences longer than `max_cha

Create {base_dir}/{ref_id}/frames/ and return the ref directory path.

Write accumulated entries as one subtitle line.

Generate a unique reference ID for a given source. Args: source: URL string, local file path, or None (for image sets). name: Opt

Resolve TTS backend with precedence: env TTS_BACKEND > user_prefs.json > 'edge'. Returns (name, source) where source is 'env', 'user_prefs', or '

Write report.json to ref_dir.

Merge override into base recursively. Override values take priority.

Return (width, height, duration, audio_codec) or None on failure.

Find the index of the last entry whose text is in punct_set.

Format seconds to SRT timestamp: HH:MM:SS,mmm

Return max chunk size for a backend.

Return duration in seconds via ffprobe. Returns None on failure.

Load and return report.json from ref_dir.

Migrate the prefs file at `prefs_path`. Returns a status dict.

Return True if this run should emit a JSON envelope on stdout.

Build backend-specific config dict from environment variables.

Compute what --delete REF_ID would change. Returns a dict suitable for both prose preview and the confirmation_required envelope's 'would_del

Return a list of structured records for every design reference. Drives both --list prose printing and the --format json envelope.

Split an oversize sentence into pieces each <= max_chars. Walks char by char; once the buffer reaches `budget = max_chars - 1`, flushes at th

Extract a short, stable ID from a video URL. Supports bilibili (BV ID), youtube watch URLs, youtu.be short links. Falls back to an 8-char md5

Print all design references with size and summary.

Migrate prefs to current version by merging missing fields from template.

Convert argparse parser actions to a JSON-serializable schema dict.

Heuristic for the 'aggressive' single-token pass. Wrap when the word is long enough and looks like a real English word rather than an abbrevi

Apply structural rewrites that deep-merge cannot do. Returns (prefs, changes) where `changes` is a list of human-readable descriptions for th

Try to load the underlying script's parser. Returns parser or None.

Scan text and return a dict of metrics + voice recommendation.

Apply SSML phoneme tags for multi-character words. Uses SAPI alphabet with numeric tones for Azure TTS compatibility.

Run the audit. Returns (issues_count, result_dict). Raises ValueError if no SECTION_CONFIG entries with `beats:` are found in the .tsx — main

Attempt to recover from known omissions. Returns list of fix descriptions.

Compute prereq state. Returns dict with backend, backend_source, required_bins, required_env_vars, missing_bins, missing_env_vars. `env` over

Copy up to MAX_FRAMES images into ref_dir/frames/. The first image is also copied as cover.<ext> in ref_dir. Returns list of copied frame pat

Classify a path or URL as 'url', 'local_video', 'image', 'unsupported', or 'not_found'. Returns one of: 'url', 'local_video', 'image', 'unsupport

Return 'horizontal', 'vertical', or 'square' based on dimensions.

Run the underlying script. Returns its exit code.

Emit JSON schema for one action, or list all actions when method is omitted.

Extract section audio from podcast_audio.wav using ffmpeg.

Extract inline phoneme markers from text: 执行器[zhí xíng qì] The regex greedily captures every Chinese character before '[', but only the last

Extract up to MAX_FRAMES representative frames from a video using ffmpeg. Builds the ffmpeg command inline (fps=1/interval, scale=-1:1080). R

Return WAV duration in seconds via ffprobe, or None if probe fails.

Write short_info.json and register_snippet.tsx for a short.

Write short_timing.json for a single-section short. Structure: - intro: INTRO_FRAMES (no audio) - content: section's original duratio

Import and return the synthesize function for a backend.

Return (width, height) via ffprobe. Returns (None, None) on failure.

Validate dependencies and env vars for a backend. Returns config dict.

Load and merge phoneme dictionaries (global + project-level) Priority (highest to lowest): 1. Explicit --phonemes argument (replaces project-

Load user_prefs.json, migrating to current version if needed.

Read podcast.txt and extract the first content line per section as its title. Returns: dict mapping section_name -> first_line_text

Read timing.json from the video directory. Raises FileNotFoundError if missing, json.JSONDecodeError if malformed. Errors propagate so main()

Manual argv slicing — argparse REMAINDER + nested subparsers misorders args (top-level parse_known_args intercepts flags as 'unknown' before t

Wrap English terms in SSML ``<lang xml:lang="en-US">`` tags by mode. Modes: ``'off'`` Return text unchanged. Use for backends that

Match section start/end times using sliding-window on word boundaries.