Voice Preservation

🌸 What Inspired Me

Last year, I lost my mom to stage 4 cancer. She was my anchor, my voice of reason, my safe place. Even though it’s been almost a year, the grief still hits like a tidal wave. There are so many moments where I wish I could just talk to her again—hear her voice, her laugh, or even just her telling me everything's going to be okay.

That’s what inspired this project. I wanted to build something that lets people like me reconnect with the voices they miss the most. Not just for nostalgia, but for healing.

What it does

Voice Preservation allows users to upload short audio clips of their deceased loved ones and use voice cloning technology to recreate a digital version of their voice. Paired with conversational AI, it gives users a way to talk to the voice of someone they’ve lost—offering comfort, connection, and maybe even a bit of closure.

For those grieving, it's not just a tool—it's a chance to hear "I love you" again, to say what they never got to say, or simply to feel close to someone who’s no longer physically there.

🛠️ How I Built It

The core of the project uses ElevenLabs for voice cloning. Users can upload short voice clips of their deceased loved ones. The system then replicates the voice and links it to a conversational AI that can chat in a tone and style similar to the person. It’s like a digital echo of someone you’ve loved and lost.

The app stack is simple but powerful:

• Frontend: Built with Vite and Tailwind CSS for a responsive, clean UI.
• Backend: Supabase
• Voice Cloning: ElevenLabs API.

** Of course, all of those was by bolt.new!

AI Chat: OpenAI’s GPT model, fine-tuned to keep the tone soft, respectful, and personal.

Challenges we ran into

• Ethical line: This was the biggest one. We didn’t want to create something dystopian or emotionally manipulative. We built with consent, transparency, and control as top priorities.
• Data Quality: Most users only have low-quality or short voice samples of loved ones. We had to work with noisy, emotional, and imperfect data.
• Emotion Management: Testing was tough. Sometimes we’d cry during demo runs. That’s when we knew—we weren’t just building a project, we were building a feeling.

Accomplishments that we're proud of

• Getting the voice clone to sound real enough to spark emotion.
• Watching teammates tear up during the first full demo—we knew then that this project meant something deeper.
• Turning raw grief into something that can help others cope, remember, and heal.

What we learned

• Grief is complex, and tech can’t solve it—but it can gently support the process.
• Building emotionally-driven AI experiences requires more than code. It requires empathy, intention, and restraint.
• Voice carries soul. Even short clips hold years of memory. Respecting that is critical.

What's next for Voice Preservation

We want to:

• Add memory-based prompt tuning—so the AI can respond with stories or quirks specific to the loved one.
• Allow families to create "memorial spaces" with voice, photos, and shared memories.
• Introduce emotion-aware conversation flows to avoid triggering responses and keep the tone compassionate.
• Work with grief counselors to ensure this platform supports healing, not dependency.

Built With

• elevenlabs
• supabase
• vite