Higgsfield AI says its Fable 5.2 model wrote a formant synthesizer from scratch, and it released a short video showing the simulated vocal tract shifting as different sounds are produced.
Higgsfield AI · Fable 5.2
An AI model says it wrote a speech synthesizer from scratch
The company released a short video pairing generated speech audio with a simulated vocal tract. The claim is unverified: the code, prompts, and speech quality have not been disclosed.
How the demonstration fits together
Step 1
Generate code
The model writes a formant synthesizer, claimed to be built entirely from scratch.
→
Step 2
Produce sound
Speech is built from resonant frequency peaks, not recorded human voices or neural networks.
→
Step 3
Show the vocal tract
A visual shows the mouth and throat shape changing from one sound to the next.
What the company claims
The synthesizer was generated entirely by the model.
Audio and visual output work together in one demo.
The clip was made with Higgsfield's own platform.
What remains open
No independent reception was found.
Source code and prompts were not released.
Speech quality and reproducibility are unknown.
Any human debugging is unclear.
Bottom line
An early showcase of AI-driven code generation for audio, not yet evidence of a new speech technology.
Release: whether the code is published.
Reproducibility: whether outside developers can run it.
Scope: whether the approach handles harder neural speech tasks.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...