Style Instructions

Control how voices deliver your content using natural language descriptions. With VoiceDesign, you describe the emotion, pace, and tone directly in the voice_description field.

How Style Works

Style control happens through the voice_description parameter in VoiceDesign. Write what you want the voice to sound like — including personality, emotion, pace, and delivery — and the model will adjust accordingly.

curl -X POST "https://api.murmr.dev/v1/voices/design" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Welcome to the future of AI.",
    "voice_description": "An excited male narrator, dramatic delivery, building anticipation",
    "language": "English"
  }' --output welcome.wav

VoiceDesign vs Saved Voices

Style instructions are part of VoiceDesign's voice_description. When you find a style you like, save the voice via the Voices API to reuse it with a stable voice_xxx ID in the /v1/audio/speech endpoint.

Style Examples

Include these descriptors in your voice_description to shape delivery. Mix and match from different categories for more specific results.

Professional

For business and formal content

calm and professionalconfident and authoritativeclear and measuredformal and precise

Emotional

Express feelings and energy

excited and enthusiasticsympathetic and caringurgent and intensejoyful and upbeat

Pace

Control speaking speed

speak slowly and deliberatelyrapid and energetic deliverynatural conversational pacethoughtful pauses between phrases

Tone

Set the overall character

warm and friendlyserious and formalplayful and lightmysterious and intriguing

Combining Styles

For more nuanced delivery, combine multiple descriptors in a single voice description. The model understands complex, layered instructions:

"A warm, friendly woman speaking at a relaxed pace, with occasional pauses for emphasis"

Combines character, tone (warm), pace (relaxed), and technique (pauses)

"Professional male narrator, approachable, clear enunciation, confident delivery"

Balances formality with accessibility

"A mysterious storyteller with a deep voice, slow build-up, dramatic pauses"

Creates narrative arc through character + delivery style

"Excited news anchor announcing breaking news, urgent but clear"

Uses role-play style instruction

Voice Descriptions by Use Case

These voice_description values combine character and style for common use cases. Try them in the Playground, then save the ones you like for reuse.

Product Demo

"A clear, professional male voice, enthusiastic but measured, highlighting key features"

Meditation Guide

"A warm, soothing female voice, calm and gentle, very slow pace, whisper-like quality"

Movie Trailer

"A deep, powerful male voice with gravitas, epic and dramatic, building intensity"

Kids Educational

"A bright, energetic young woman, playful and animated, speaking clearly, fun and engaging"

Podcast Intro

"A friendly, conversational male voice, natural delivery, welcoming host energy"

Tips for Effective Descriptions

Be descriptive, not prescriptive

Describe the feeling you want ("warm and inviting") rather than technical instructions ("lower pitch by 10%").

Include both character and delivery

The best descriptions combine who the speaker is ("elderly wizard") with how they speak ("slowly and deliberately"). See the Voice Crafting Guide for more patterns.

Keep it under 500 characters

2-3 descriptors usually work better than long, complex instructions. Focus on the most important characteristics.

Test in the Playground

Experiment with different descriptions in the Voice Playground before committing to your application code.

Save and Reuse Voices

Once you find a voice description that captures the right style, save it as a reusable voice. This gives you a consistent voice_xxx ID for production use:

Workflow

1. Design a voice with style descriptors:
   voice_description = "A calm, professional narrator, measured pace, clear enunciation"

2. Generate and preview audio in the Playground

3. Save the voice → get voice_abc123

4. Use in production with /v1/audio/speech:
   { "text": "Your content here", "voice": "voice_abc123" }

Saved voices capture the style from your description — every generation with that voice ID will have the same character and delivery.