I released Inflect v3: lightweight local TTS in 34M parameters (Apache 2.0)

·2026.11.12 10:36

Key point

The release includes Mini, Micro, and Atom variants, with the 34.31M parameter Mini model supporting reference voice tuning and 48 kHz enhancement on CPU and CUDA.

Details

Inflect v3 is a family of local text-to-speech models released under the Apache 2.0 license, comprising Mini, Micro, and Atom variants.
The Mini model features 34.31M parameters, six voices, and runs on both CPU and CUDA, while Micro has 11.79M parameters and Atom has 157K parameters.
Mini supports reference-voice tuning to adjust speech toward a provided recording and includes vocal-event tags for sounds like [laugh], [sigh], and [gasp].
Vocal-event tags currently use repeatable performances with limited variation and do not affect the speaking style of surrounding words.
Weights and inference code are available on GitHub and Hugging Face, with third-party components retaining their own licenses.

open-source local-llm text-to-speech speech-synthesis apache-2.0 inflect

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.