July 25, 2026
Small model, big mouth
Inflect-Micro-v2: complete voice in 9.36M parameters
Tiny voice maker stuns fans, but the comments instantly demand cloning and more voices
TLDR: Inflect-Micro-v2 is a very small offline tool that turns text into speech and has impressed many readers with how good it sounds for its size. But the comments quickly split between hype and reality checks, with fans praising it and others complaining that one voice, English-only output, and no voice cloning limit the wow factor.
A teeny speech generator called Inflect-Micro-v2 just dropped, and the crowd is basically doing a double take: you’re telling me this little thing can turn text into spoken audio on your own machine, without needing the cloud, and it still sounds this good? That was the vibe across the thread, with multiple commenters openly shocked by how much voice quality Owen squeezed into something so small. One fan called it “amazing quality for such small size,” while another said it was so good they immediately swapped out their older setup and even posted a working implementation. That’s not just praise — that’s the open-source version of moving in together.
But of course, the comments didn’t stay wholesome for long. The biggest mini-drama? Expectation vs. reality. One reader had to step in and clarify that “complete voice” does not mean a full speak-and-listen package, and definitely does not mean celebrity-style voice copying. In plain English: it reads text out loud, but it’s English-only, one fixed male voice, no cloning. That immediately became the thread’s “record scratch” moment, especially since another commenter jumped in with the wish heard round the replies: “I wish there were a voice clone option.”
And then came the snark. Not everyone was ready to crown it king. One blunt hot take said the sound is impressive for the size, but still not especially pleasant to hear, comparing it to older text-to-speech tools. So the verdict from the crowd? A small miracle, a practical win, and also not quite the silky robot dream some people wanted.
Key Points
- •Inflect-Micro-v2 is a local English text-to-waveform TTS model with 9,356,513 deployable parameters, 37.53 MB FP32 size, and 24 kHz mono output.
- •The Inflect v2 release exposes one public API across two model sizes: Micro for quality under 10M parameters and Nano for footprint under 4M.
- •The article reports separate evaluation dimensions rather than one overall score, including human preference, UTMOS22 predicted naturalness, multi-ASR intelligibility, package footprint, and runtime.
- •Inflect-Micro-v2 achieved a 66.2% preference rate in an anonymous community study and a 4.395 UTMOS22 score with a 95% bootstrap confidence interval of 4.381–4.408.
- •In CPU benchmarking on a Hugging Face instance, Inflect-Micro-v2 ran at 6.28× real-time and Inflect-Nano-v2 ran at 10.72× real-time.