Skip to content
ThreadOut
Václav Volhejn@vvolhejn · Aug 17, 2026

New side project: Samuel, a model I trained to mimic speech using Pink Trombone, a (pre-existing) silly vocal tract…

3 tweets1 min read♥ 2.8Koriginal

Summary

A machine learning researcher created Samuel, a speech autoencoder that uses Pink Trombone (a vocal tract simulator) as its decoder to generate intelligible speech. The project is non-trivial because Pink Trombone is non-differentiable and sample-by-sample, requiring careful optimization to maintain intelligibility rather than just matching waveforms.

Summarized by ThreadOut AI from the full thread. May miss nuance — read the thread below.

  1. #1

    New side project: Samuel, a model I trained to mimic speech using Pink Trombone, a (pre-existing) silly vocal tract simulator. 🧵

  2. #2

    From an ML perspective, it's a speech autoencoder where the decoder is Pink Trombone. It's a non-trivial task: Pink Trombone is non-differentiable, sample-by-sample, and if you optimize the model to just match the waveform, it's not intelligible. Write-up coming soon!

  3. #3

    Try it: samuel.vvolhejn.com Credits to Neil Thapen for the original Pink Trombone and @ConcreteSciFi for the programmable version I used as the starting point.