🗣️ Inflect v3 is now on Pinokio: 1-click install
Inflect v3 is Owen Song's family of compact English text-to-speech models, from 34.3M parameters down to just 157K.
The launcher runs the official Inflect v3 playground on your own machine. It uses your NVIDIA GPU when you have one
and falls back to CPU otherwise.
🎧 What it does
🧠 Includes all 3 models: Mini (34.3M, best quality), Micro (11.8M) and Atom (157K).
🎙️ Comes with 6 voices: cassidy, mark, hope, james, sophia and jasper.
😂 Mini understands vocal-event tags: [laugh] [sigh] [gasp] [cough] [yawn] [inhale] [exhale] [throat_clear] (up to 4
per take).
🎚️ Mini can also output 48 kHz audio through an optional enhancement.
⚖️ Has 3 modes: Create (one model), Compare (all three models on the same text) and Quality (Mini at 24 kHz vs 48
kHz).
🔌 Includes a local Gradio API at /synthesize for your own apps and scripts (Python, JS and curl examples are in the
README).
💬 Example
You really surprised me. [laugh] That was a good one.
📋 Before you install
💻 English only. Each take is limited to 800 characters.
📦 On the first start, all three models download automatically. After that it runs offline.
🗑️ Generated takes are deleted after an hour, so download the ones you want to keep.
📄 The Inflect code is Apache 2.0. The v3 weights use the Inflect preview evaluation license:
huggingface.co/owensong/Inflect…
👉 Install → Start → Generate
github.com/PierrunoYT/Inflect-v…
#inflect
Browse tags
Most posts · Last 30 days
No recent tagged posts yet.
You’re all caught up.
Join the community
Follow people and apps for updates. Explore Announcements and All for more.
Discover people →
