Who it is forA Persona Voice user who speaks to their assistant in Japanese, Chinese or English and wants the best quality and latency.
The problemOne voice model does not fit every language, and picking wrong gives worse accent, quality or delay.
The outcomeThe model that fits your language and hardware, with streaming block size and speed you understand before you commit.
What you need first
- Apple silicon Mac or x64 machine with NVIDIA CUDA
- bun to run from source
Steps
- Install Persona Voice and start first-run setup
- For English and most other languages, download Chatterbox (640 ms streaming blocks on Apple Silicon)
- For Japanese or Chinese, download Seed-VC Tiny (300 ms streaming blocks)
- Add the other model later in Settings, Voice model
- Compare both on the same recording; on an M4 Pro Chatterbox used 29% less processing time in English
- Check docs/VOICE_MODELS.md for startup delay and storage before downloading both
Prompt for your agent
I use Persona Voice mostly in <language> on <hardware>. Based on the README's voice model table, which model should I download first, what latency should I expect, and how do I switch later?
Caveat. Quality varies by language and voice.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for miuuyy/persona-voice and "Choose the right voice model for your language"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer