SuperAPI AI Directory

audio-generation

General & other

View repository

Updated 10-16 09:58

About this skill

Audio Generation

Route First

User need Read next
Choose a TTS engine — compare backends, pick based on hardware/needs references/backends.md
Turn any text into spoken audio — single voice, short or long references/text-to-audio.md
Produce a multi-voice podcast episode from a script or source material references/podcast.md (technical) + references/podcast-principles.md (editorial)
Convert an article, document, or long-form text into narrated audio references/long-reads.md (technical) + references/long-read-principles.md (editorial)
Audio assembly, concatenation, normalization, fades, format conversion references/ffmpeg-audio.md
Choose a voice — catalog, language coverage, style matching references/voices.md

If the request is ambiguous, ask: "Is this a single voice reading, a multi-person conversation, or a narration of existing content?"

Mental Model

  • TTS backends are pluggable. Pick the right one for the user's environment — see references/backends.md for the decision flowchart. Options range from cloud (edge-tts) to local CPU (Kokoro, Piper, MeloTTS, Pocket TTS).
  • ffmpeg is the audio workbench. Concatenation, trimming, format conversion, fades, mixing, normalization — all post-production lives here.
  • pydub (optional) is for Python-level audio control when ffmpeg CLI isn't enough.

Core Rules

  1. Pick the right backend. Don't default blindly — check the user's hardware, offline needs, and voice cloning requirements. See references/backends.md.
  2. Normalize loudness on every final deliverable (ffmpeg -af loudnorm).
  3. Default output: MP3 at 24kHz mono. Use WAV only if the user needs lossless or the backend outputs WAV natively.
  4. Chunk long text at paragraph boundaries, never mid-sentence. Chunk size depends on backend (see backend-specific docs).
  5. Name files descriptively: podcast-episode-1.mp3, article-narration.mp3, intro-audio.mp3.

Dependencies

  • Required: one TTS backend (see references/backends.md), ffmpeg (system install)
  • Optional: pydub (pip install pydub), markitdown (pip install markitdown)