I bootstrapped a foundational text-to-speech model from scratch

Name: I bootstrapped a foundational text-to-speech model from scratch
Availability: InStock
Author: vincenttjona

by vincenttjona·Apr 8, 2026·3 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidBold BetBig Brain

Building a foundational TTS model from scratch is rare — ElevenLabs charges 10× more.

Strengths

•Actual model training from scratch, not an API wrapper — genuine ML infrastructure work.
•200ms latency and $5/million chars undercuts ElevenLabs pricing significantly.
•Mobile apps already shipped with thousands of users — not just an API demo.

Weaknesses

•Quality claims need independent verification against ElevenLabs and PlayHT benchmarks.
•Enterprise TTS is well-funded; competing on price alone may not sustain the business.

Post Description

After being frustrated by the current text-to-speech models in terms of naturalness and pricing, my brother and I built our own model from scratch. We just launched our API. Pricing is $5 / Million Chars and Latency ~200ms for enterprise customers. Appreciate your feedback.