← All posts

The Best Podcast and Audio Skills for Claude Code

2026-10-17

Audio skills are where agent workflows stop being text and start costing real money per minute. The scores here reflect both output quality and how well each skill controls the expensive part.

Seven skills in the directory produce or process audio. They cover four jobs: speech synthesis, music, audio-driven video, and transcription.

Speech synthesis

doubao-tts — 9.1 · Productivity

Volcengine Doubao Speech 2.0 for Mandarin and multilingual narration. The highest-scoring TTS skill for Chinese, and Chinese prosody is where most Western TTS models still sound wrong.

fish-audio-tts — 8.9 · Productivity

Expressive multilingual narration via fish.audio S1/S2. Choose it over Doubao for expressiveness and language coverage; choose Doubao for Mandarin naturalness.

voice-dossier-builder — 8.7 · Marketing

Codifies how a founder or brand sounds — vocabulary, rhythm, and phrasing — so generated narration matches the person. Useful precisely because TTS can produce fluent speech that sounds nothing like the speaker.

Music

music — 8.5 · Developer Tools

Music generation via the ElevenLabs Music API — instrumental tracks and backgrounds. Straightforward API work, competently done.

music-to-video — 9.4 · Video & Motion

Turns an audio track into a video, or pulls audio from a video to re-cut. The highest score in the group, and the one with the most complex pipeline — audio analysis, timing, and rendering all have to agree.

Video and transcription

video-voiceover — 8.6 · Video & Motion

Voiceover for video production: narration timed against existing cuts.

baoyu-youtube-transcript — 9.3 · Writing

Downloads transcripts and subtitles, plus cover images. The cheapest entry point into audio workflows — and often the highest value, because the transcript is the input to everything else.

The cost shape is different

Text tokens are cheap and predictable. Audio is neither. A minute of generated speech or a rendered music track costs orders of magnitude more than a paragraph of text, and the failure mode is discovering it only at the render step.

Two practical consequences:

  1. Test the pipeline with the shortest possible clip. A 10-second render tells you whether the voice, timing, and mixing are right.
  2. Separate draft and final. Generate a cheap draft, review it, and only then render at full quality — music-to-video and the TTS skills all benefit from this split.

    3. How to choose

    • Chinese narration? doubao-tts.
    • Expressive, multilingual? fish-audio-tts.
    • Brand voice consistency? Add voice-dossier-builder.
    • Audio into video? music-to-video or video-voiceover.
    • Just need the words? baoyu-youtube-transcript.

    The video category covers the rest of the production pipeline.