Proxy that converts LaTex notation into spoken language

[HanFarJR]

Why I made it

Kokoro-FastAPI has a specific gap in their utility. They read LaTeX markup character by character rather than converting it to spoken language. When encountering math notation, the TTS engine will hear "dollar sign backslash frac open brace…" instead of "x squared plus one over two". My tts-latex-proxy solves this problem.

Kokoro-FastAPI and other TTS-engines are rendered useless for scientific notation without a preprocessing layer that converts said notation to spoken English like so:

What it does

Open WebUI (:3000) → tts-latex-proxy (:8881) → Kokoro-FastAPI (:8880)

The proxy intercepts /v1/audio/speech requests, processes the input text, then forwards the cleaned version to your TTS backend. All other API calls (models, health checks) pass through transparently.

Before / After

Raw LLM output	What the TTS engine receives
$\frac{a}{b}$	a over b
$E = mc^2$	E equals m c squared
$\int_0^1 f(x) dx$	the integral from 0 to 1 of f of x d x
$\alpha + \beta$	alpha plus beta
$\sqrt{x^2 + y^2}$	the square root of x squared plus y squared
$\sum_{i=1}^{n} x_i$	the sum from i equals 1 to n of x sub i
$P(A \cup B)$	P of open paren A union B close paren

Markdown formatting (**bold**, # Headers, [links](url), bullet lists, tables) is stripped silently.