Gen3 Admin Model Guides Orpheus V1 English

Orpheus v1 English — Vocal Directions

Start Here

1. On Models → Inference Providers, confirm Groq is connected and discovery succeeds.
2. On Configured Models, enable canopylabs/orpheus-v1-english (import from Groq if missing).
3. On Default Models, set Default TTS to Orpheus v1 English when this is the deployment-wide speech default.
4. On Platform Multimodal Settings, enable Text-to-speech (TTS) so tenant chat shows Read aloud and Hands-Free reply audio.
5. Tell tenant users and agent builders: the chat model must include bracketed vocal directions in the visible reply (for example [cheerful]) before Read aloud—GT AI OS does not inject directions after generation.

Why this matters

Canopy Labs Orpheus v1 English is an expressive Groq TTS model. Bracketed hints such as [cheerful], [whisper], or [professionally] control delivery. Operators enable the model and TTS gates; tenants and agents supply the bracket syntax in message text.

Details

Operator setup checklist

Control Panel tab	Action
Inference Providers	Groq API key valid; canopylabs/orpheus-v1-english discoverable
Configured Models	Orpheus row enabled for TTS
Default Models	Default TTS → Orpheus when appropriate
Platform Multimodal Settings	TTS enabled (disabled TTS hides Read aloud in tenant chat)

After platform setup, tenants configure per-agent Speech Settings (TTS model and voice) on Building Agents. Universal baseline free-tier flows often set Orpheus as default TTS after the Groq API key step in QuickStart.

How vocal directions work (Groq / Orpheus)

Per Groq's Orpheus documentation:

Pattern	Effect
No brackets	Natural, conversational delivery
One or two word directions in [brackets]	Subtle style—[cheerful], [whisper], [professionally]
Multiple directions in one line	More expressive, acted delivery

Provider best practices: use 1–2 word bracket tags; omit brackets for neutral cadence; keep each Groq TTS input under 200 characters; spell numbers with hyphens when letter-by-letter reading matters (2-0-3).

Tenant and agent workflow (support copy)

GT Chat Read aloud synthesizes the visible assistant message text. The chat LLM does not add Orpheus brackets unless prompted.

Per-message: users ask the agent to prefix spoken lines with a direction—for example "Start with [cheerful] then summarize status."

Agent system prompt: builders add standing guidance on Building Agents → Speech Settings or instructions—for example "When Read aloud may be used, prefix concise answers with one Orpheus direction in square brackets."

Verify before listen: users confirm [direction] tokens appear in the message before Read aloud.

Length limits

Groq documents 200 characters per Orpheus request. Long assistant answers may be chunked for synthesis; very long Read aloud runs can fail or sound uneven. Recommend short spoken summaries with inline directions for reliable expressive delivery.

Example (tenant GT Chat)

1. Agent has Orpheus on Speech Settings.
2. User: "In one sentence, say whether backups finished. Start with [cheerful]."
3. Assistant: [cheerful] Backups completed successfully at 02:15 UTC.
4. User clicks Read aloud.

Related pages

• Model Guides
• Models
• Model Cards — Groq Orpheus
• QuickStart Setup Guide
• Tenant copy: Orpheus v1 English — Vocal Directions, Accessibility