Seeking Guidance on Creating a Realistic Santa Claus Voice Using Gemini TTS Voice Design

Hi everyone,

I’m trying to create a realistic Santa Claus voice using Gemini TTS Voice Design, but haven’t been able to achieve the desired results despite multiple iterations.

I’m looking for a deep, elderly male voice with a naturally raspy and gravelly texture, a warm and jolly grandfatherly personality, and expressive storytelling delivery.

My approach so far:

I’ve provided reference audio recordings to Gemini 4 Argon to analyze the voice characteristics, including pitch, resonance, vocal texture, age and speaking style. I then used the extracted acoustic properties, along with Claude, to formulate detailed voice-design prompts.

Despite trying multiple prompt variations, the generated voices still sound like generic elderly male voices rather than the Santa Claus character I’m aiming for.

For reference:

  • Voice Design Prompts – Document

  • Generated Audio Samples

The document includes the different prompt iterations I’ve tried, and the Drive folder contains their corresponding audio outputs.

Could anyone from the Google team or community suggest better prompting techniques, recommended approaches, or whether voice cloning would be more suitable for this use case?

Any guidance would be appreciated. Thanks!

Hello @suraj_kumar9,

Here are a few tips for getting a more distinct character voice with Gemini TTS:

  • In Voice design, keep the voice description short and specific (1–2 clear sentences) and limit it to permanent vocal traits such as age, timbre, vocal texture, and accent. Long or overlapping descriptions tend to produce more generic, less consistent results.
  • Put delivery and personality (for example “warm, jolly storytelling”) in a short style prompt at synthesis time, and use inline tags such as <laugh> or <short pause> in the transcript for moments like a hearty laugh. See the TTS prompting guide.
  • If Voice design still doesn’t capture the exact timbre, Voice replication can create a voice from a 10–30 second reference clip. It requires a consent recording from the same adult speaker, so it works only if you have a voice actor who can record both clips.