Piper AI Voice Clone training
Budget / Salary$10–30
TypeFreelance project
LocationRemote
Posted50 minutes ago
***** MUST WORK WITH MY LIMITED BUDGET OF MAX $30 USD, SERIOUS & HONEST FREELANCERS ONLY, YOU MUST HAVE EXPERIENCE WITH AI VOICE CLONING AND PIPER **
This project is less than 12 hour job for Piper AI Clone training. Freelancers don't complain or make excuse about the cost of GPU. We can pay you extra for GPU or we can spin up a GPU server for you. Or if you have your own PC with a GPU to use for this project let us know. We are a radio station and our funds are tight. We also only less 3 to 12 hours max so many web hosts will offer free GPU limited servers for the 12 hours like kaggle.com or google colab
I have a library of 100+ mp3 recordings of one speaker. YOU DO NOT NEED TO SPLIT - CONVERT ALL 100 + FILES. I JUST WANTED TO PROVIDE ENOUGH FILES TO HELP YOU. PICK ONES U WANT TO WORK WITH. SO YOU MAY ONLY CHOOSE LESS THAN 30.
You may need to strip background bed music using demcus. The music has a low DB and dry voice is louder than music bed. And music does go away or is not noticable as time goes on. So voice is dry and loud and cleae. All 100 plus files are under 5 min each.
Your NOT converting all 100 + files. I just want to provide you enough files so you can clone ai voice. Pick as little or as many files as you want to work with.
The goal is a voice that sounds indistinguishable from the original; I want an exact match because the model will power an on-air AI DJ for a radio stream. You can bootstrap with the piper “ryan” preset, but the finished model must faithfully reproduce my speaker’s timbre, pacing, and articulation under normal broadcast processing.
You’ll handle every step on your own GPU rig. When you’re done, hand over the compiled ONNX files (and any supporting config/json) ready to drop into Piper-TTS.
Deliverables
• Cleaned & aligned training dataset
• Trained Piper model in ONNX format (+ metadata)
• Demo audio proving voice accuracy
This project is less than 12 hour job for Piper AI Clone training. Freelancers don't complain or make excuse about the cost of GPU. We can pay you extra for GPU or we can spin up a GPU server for you. Or if you have your own PC with a GPU to use for this project let us know. We are a radio station and our funds are tight. We also only less 3 to 12 hours max so many web hosts will offer free GPU limited servers for the 12 hours like kaggle.com or google colab
I have a library of 100+ mp3 recordings of one speaker. YOU DO NOT NEED TO SPLIT - CONVERT ALL 100 + FILES. I JUST WANTED TO PROVIDE ENOUGH FILES TO HELP YOU. PICK ONES U WANT TO WORK WITH. SO YOU MAY ONLY CHOOSE LESS THAN 30.
You may need to strip background bed music using demcus. The music has a low DB and dry voice is louder than music bed. And music does go away or is not noticable as time goes on. So voice is dry and loud and cleae. All 100 plus files are under 5 min each.
Your NOT converting all 100 + files. I just want to provide you enough files so you can clone ai voice. Pick as little or as many files as you want to work with.
The goal is a voice that sounds indistinguishable from the original; I want an exact match because the model will power an on-air AI DJ for a radio stream. You can bootstrap with the piper “ryan” preset, but the finished model must faithfully reproduce my speaker’s timbre, pacing, and articulation under normal broadcast processing.
You’ll handle every step on your own GPU rig. When you’re done, hand over the compiled ONNX files (and any supporting config/json) ready to drop into Piper-TTS.
Deliverables
• Cleaned & aligned training dataset
• Trained Piper model in ONNX format (+ metadata)
• Demo audio proving voice accuracy
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.