nc://podcast

Podcast Studio: turns a two-host script into a mastered MP3 episode, one Microsoft Edge voice per host. A Nodecode rewrite of suhas-svg/hermes_podcast by suhas-svg.

Get it

Tell the agent "add the podcast cell", or run nodecode add nc://podcast.

Set it up

Nothing is required: the voice service (Microsoft Edge's read-aloud voices) needs no key, and ffmpeg is used when it is on PATH. The podcast section has defaults for both hosts' voices and the pause between turns:

{
  "podcast": {
    "host_a_voice": "en-US-AndrewMultilingualNeural",
    "host_b_voice": "en-US-AvaMultilingualNeural",
    "pause_ms": 350,
    "intro_chime": true,
    "episodes_dir": "~/podcast/",
    "ffmpeg_path": ""
  }
}

Without ffmpeg the episode is still produced: the clips are joined as they are, with no chime, pauses, loudness pass or embedded chapters, and the answer says so.

Use it

Ask for what you want to listen to: "make a 4-minute podcast about this paper" or "a daily briefing on AI research". The model reads the sources, writes the script and calls:

• podcast_produce — title, description, hosts (1-3) and lines (speaker, text, optional chapter); answers the episode id, duration, chapters, the MP3 and its timestamped transcript;
• podcast_voices — the voice list for a locale, and gender;
• podcast_episodes — what was produced, newest first, or one episode again by id.

/podcast lists episodes, /podcast voices es lists Spanish voices and /podcast doctor says what this machine has. (help :podcast) carries the writing guidance the original shipped as a skill: read the sources first, two hosts, 150 spoken words a minute, no markdown read aloud.

Every episode is a folder under the episodes directory: episode.mp3, transcript.md with a timestamp per line, and episode.json with the hosts, chapters and duration.