- Home
- Skills
- Productivity
- muse-voice-recipe
Muse Voice Recipe
Design your AI agent's own voice from 3 questions, screen 16 candidates automatically, pick the winner in one audition file, and freeze it as a Voice Recipe Card anyone can reproduce.
Free
Works with the AI Agents you already use
muse-voice-recipe
Example session with this skill installed
I need a voice for a helpful assistant. She should be a young woman with a medium pitch, gentle and soft brightness, and an earnest, hardworking speaking style. Rank 10 candidates for me.
- Read your context and instructions
- Compiled the muse-voice-recipe
I have synthesized the voice caption and screened the initial candidates. The caption has been locked to ensure consistency across the top seeds.
Voice Design Results: Assistant-YF-M
| Metric | Value / Status |
|---|---|
| Generated Caption | "Young female voice, medium pitch, gentle and soft, speaking in an earnest and hardworking manner." |
| Mean F0 (Pitch) | 302 Hz for the voice chosen by listening (candidates ranged 251–325 Hz) |
| Primary Screening | 10 candidates screened; the top 5 were put into one audition file. |
| Audition File | outputs/audition_yf_m.wav (5 candidates separated by beeps) |
Top 5 candidates
The top 5 candidates from the first screening were stitched into one audition file with beep separators. Listening to that single file, the voice with a mean F0 of 302 Hz was chosen — even though it ranked 8th out of 10 in the automated first screening, the listening test was given priority. The seed is fixed and recorded in the Voice Recipe Card.
What happened next
- The audition file was listened to and the voice with a mean F0 of 302 Hz was chosen.
- The 3-emotion test (joy, sorrow, neutral) was run before the voice was fixed.
- Call
make_recipe_card.pyto freeze the chosen voice and issue the Voice Recipe Card (PNG + recipe.yaml).
Connects securely to your tools. The creator never sees your data.
About this skill
Most agent voices are found by luck: generate, listen, regenerate. This skill turns voice design into a repeatable procedure.
You start from the voice you imagine. Three questions — voice type and pitch, brightness, speaking style — are composed into a precise voice-design caption. The skill then generates 16 seed candidates and screens them by pitch, voiced ratio, and clarity, so you only listen to the top 5, stitched into a single audition file with beep separators. One listen, one number, done.
Before the voice is final, a 3-emotion test (joy, sorrow, neutral) checks that the same voice can actually act. The result is frozen as a Voice Recipe Card (PNG + recipe.yaml): caption, seed, engine settings, and reference settings in one place, so the same voice can be reproduced later or in another environment — and changed deliberately, by issuing a new card, never by accident.
The method was developed and used in production to design a working agent persona's voice end to end.
What is included:
- SKILL.md with the full 5-step procedure
- compose_caption.py — build the design caption from 3 answers
- screen_candidates.py — automatic first screening of candidates
- make_audition.py — one-file audition of the top candidates
- make_recipe_card.py — issue the Voice Recipe Card (PNG + YAML)
Requirements:
- The engine skill muse-irodori-voice (free, MIT, on GitHub) installed and working; it provides Irodori-TTS synthesis
- Model weights are not included; you obtain them from the original distributors, as their licenses require
Important notes:
- This skill designs original voices from trait descriptions. It is not a tool for imitating or cloning the voice of a real person (including voice actors and public figures) or a copyrighted character, and such use is against its rules.
- If you publish audio made with a voice designed by this skill, please disclose that it is an AI-generated voice.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Productivity.
Requires muse-irodori-voice setup. Does not support character or actor names for voice generation. Final selection requires manual human listening.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 9 days ago
- Free with an Agensi account
- Software
- muse-irodori-voice
Free on GitHub; it provides Irodori-TTS synthesis.