Voice

Dyva includes a library of professionally designed preset voices that cover a wide range of ages, genders, accents, and energy levels. You can preview any preset by tapping the play button next to it in the voice selector. Each preview reads a sample sentence so you can hear how the voice sounds in context.

Presets are grouped into categories to make browsing easier: warm, authoritative, energetic, calm, youthful, mature, and more. Each voice has been tuned for natural-sounding conversation — they handle questions, exclamations, pauses, and emotional shifts well.

Choosing a preset takes seconds, and you can change it at any time without affecting existing conversations. If you're unsure which voice fits, try a few: create a test chat, enable voice playback, and listen to how each option delivers your character's greeting and typical responses. The right voice will click immediately — you'll hear it and think, "That's them."

If no preset captures the voice you hear in your head, you can create a custom voice using voice cloning. Upload one or more audio samples (at least 30 seconds of clear speech, ideally 1-3 minutes), and the system will generate a voice model that closely matches the speaker's tone, pitch, pacing, and cadence.

For the best results, follow these guidelines when recording samples:

• Use clean audio. Record in a quiet environment with minimal background noise. A good microphone helps, but even a phone in a quiet room works.
• Speak naturally. Don't read in a monotone or over-enunciate. Talk the way you'd talk in a real conversation — the AI needs to hear natural speech patterns to reproduce them.
• Show range. Include questions, statements, and emotional variation in your sample. If the character needs to express excitement, record some excited speech. This helps the voice model handle different contexts.
• Avoid music or effects. The sample should be voice only. Background music, sound effects, or other speakers will confuse the cloning model.

Cloned voices are private to your account. You can use them across multiple characters. Please only clone voices you have the right to use — your own voice or voices you've obtained explicit permission to clone.

Voice calls let users have a live, real-time spoken conversation with your character. Instead of typing messages and reading responses, they talk and listen — just like a phone call. The character hears the user's voice (converted to text via speech recognition), generates a response, and speaks it back using its assigned voice.

For users, voice calls feel remarkably natural. The character responds with appropriate pacing, handles interruptions, and maintains the same personality it has in text chat. For creators, voice calls unlock a whole new dimension of engagement — especially for characters designed around companionship, coaching, tutoring, or storytelling.

To enable voice calls for your character, simply assign a voice (preset or cloned). Voice calls are then automatically available to users who interact with your character. There are no additional settings to configure — if the character has a voice, it can take calls.

Keep in mind that voice conversations tend to be shorter and more spontaneous than text chats. Your character's definition should work well in both modes. If your character relies heavily on formatting (like bullet lists or code blocks), consider adding a note in the definition: "In voice conversations, present information conversationally rather than in lists."

Beyond choosing a voice, you have several controls to fine-tune how your character sounds:

• Speed — Adjust how fast the character speaks. A fast-talking auctioneer character might use a higher speed, while a meditative guide would use a slower pace. The default is 1.0x, and you can adjust from 0.5x to 2.0x.
• Stability — Controls how consistent the voice sounds across different messages. Higher stability means more predictable delivery; lower stability introduces more expressive variation. For most characters, the default works well. Lower it for characters that should sound more emotional or animated.
• Clarity — Enhances the crispness and articulation of the voice. Turn this up for characters that need to be clearly understood (like a teacher or narrator) and leave it at default for more casual, naturalistic delivery.

These settings affect all voice interactions — both voice calls and text-to-speech playback. Experiment with different combinations to find the sweet spot for your character. Small adjustments can make a big difference: dropping the speed by 0.1x and lowering stability slightly can turn a generic-sounding voice into something that feels distinctly human.