Hermes Audio Drama

Blind-first creative tooling for audio storytelling with Hermes Agent.

This repository contains:

• a reusable Hermes skill for multi-voice audio drama production,
• a manifest-driven renderer based on Edge TTS + synthetic ambience,
• a finished English demo audio drama: "Aishka and the Blog".

What it does

The workflow turns a story brief into a short audio drama with:

• multiple TTS voices,
• explicit speaker switching,
• timed background sound cues,
• final WAV/MP3 export,
• a render manifest with scene and line timings.

The first included demo is based on Denis building his blog together with Aishka, his AI companion.

Included demo

• Final audio: build/final_mix.mp3
• Render metadata: build/render_manifest.json
• Story manifest: assets/story.yaml
• Voice mapping: assets/voices.yaml

Repository structure

assets/
  story.yaml
  voices.yaml
build/
  final_mix.mp3
  render_manifest.json
docs/
  concept.md
scripts/
  render_audio_drama.py
skill/
  audio-drama-production/
    SKILL.md
    templates/
    references/
plan.md
README.md

Requirements

Python packages:

• numpy
• PyYAML
• edge-tts

System tools:

• ffmpeg
• ffprobe

Installation

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Make sure ffmpeg and ffprobe are available in PATH.

Render the demo

python3 scripts/render_audio_drama.py

Expected output:

• build/final_mix.wav
• build/final_mix.mp3
• build/render_manifest.json

How the pipeline works

1. Read assets/story.yaml and assets/voices.yaml
2. Render each spoken line with Edge TTS
3. Convert lines to a common mono WAV format
4. Synthesize stylized ambience and one-shot cues in Python
5. Mix scene-by-scene on a shared timeline
6. Export WAV + MP3 and save timing metadata

Voice casting in the demo

• Denis: en-US-AndrewNeural
• Aishka: en-US-AvaNeural

Aishka is intentionally voiced as female.

Why this matters

Most creative-agent demos are visual-first. This project explores an alternative:

text-driven, audio-first, accessible creative tooling.

That makes it a fit for:

• audio storytelling,
• creative agent workflows,
• accessibility-focused creative software,
• hackathon demos where the tool is part of the creative output.

Hermes skill

The reusable Hermes skill lives in:

• skill/audio-drama-production/SKILL.md

It includes templates and notes for creating new short audio dramas with the same workflow.

Notes

• Background sounds are synthetic and stylized by design.
• The goal is readability and atmosphere, not full cinematic realism.
• The first demo is intentionally short and hackathon-friendly.

License

MIT