Six years in, VolceMood is still nineteen people arguing about one number: added latency.
Every competitor's homepage leads with a voice count. Ours leads with a stopwatch, because the number that decides whether a voice changer survives a ranked match or a live stream is milliseconds, not menu items.
19 people, four functions
v5, six years of revisions
under 20 ms, input to output
The short version
VolceMood started as an attempt to fix one specific complaint: every real-time voice changer on the market added enough delay that a streamer could hear their own voice arrive late in their headset, a fraction of a second behind their own mouth. That fraction is the whole product now.
We did not set out to build the biggest voice-changer library on the market, and we still haven't. Twelve tuned voices ship in the Creator tier, the same twelve in Streamer Pro, because a voice that glitches at normal speaking volume is worse than no voice at all — a competitor's ninety "voices" are mostly the same three pitch shifts renamed and re-skinned, and testing each new one against real speech at real volume is slower than adding it, which is exactly why we do it anyway. The engine that resulted, version 5 as of this year, holds under 20 milliseconds of added latency measured input to output on a mid-range laptop — not a studio machine, an ordinary five-year-old laptop someone actually streams from.
The engine got faster before it got bigger.
A pitch shifter that missed the point
The first build did one thing: shift pitch up or down. It worked, and it sounded exactly as artificial as every other pitch shifter on the market, because pitch alone can't fake a different voice — formant has to move too, and it has to move on its own dial or the result sounds like a chipmunk or a demon, not a person.
The two-dial model and the first latency budget
Pitch and formant split into two independently addressable controls. At the same time, the team set a hard internal ceiling on added latency instead of chasing it opportunistically — the number that ceiling produced is the one still quoted today. Voices that couldn't hit it were cut, not shipped with a warning.
Soundboard, noise gate, and platform routing
The hotkey soundboard and the noise gate arrived once the core engine was stable enough that adding features didn't reopen the latency question every time. Routing into OBS and Discord at once — now Streamer Pro's defining feature — took another full revision to get right without doubling the processing load.
Version 5, a metered API, and a team that hasn't stopped arguing about the number
The current engine ships across desktop, mobile and now a metered REST endpoint for developers who don't want a client at all. Nineteen people work on it. Eleven of them still spend most of their week on the audio pipeline itself, which is roughly the point.
Where the nineteen people actually sit.
Most of the team works on the thing that's hardest to fake: the audio pipeline. The rest keep the clients current and the tickets answered.
Latency-first, stated plainly
We will turn down a feature that breaks the latency budget.
That sentence isn't a value statement, it's an operating rule. If a proposed voice, effect or routing option adds enough processing time to push the engine past 20 milliseconds of added latency on ordinary hardware, it doesn't ship — not "ships behind a warning," not "ships and gets fixed later." The two-dial pitch/formant model exists because the alternative, one combined dial, produces a voice that sounds obviously fake the moment someone speaks at normal volume, and a fake-sounding voice changer for gaming or streaming is worse than an honest, smaller one.
This is also why there's no cloud round-trip on the desktop or mobile engine. Processing on-device is what keeps the number under 20 ms in the first place — a server call, even a fast one, adds network time on top of processing time, which is why the Developer API tier is the one exception: an API is a network call by definition, and we say so on the plans page rather than quietly hoping nobody does the math.
None of this comes with a certification badge, because none has been issued to VolceMood and we're not going to print one that implies otherwise. What we'll commit to instead is a working refund: fourteen days from purchase, if the engine won't hold sync on the customer's own hardware, checked against an actual support ticket rather than taken on faith.
Who's actually running this.
Volcemood trades as VolceMood at volcemood.info. It's a software company, not a storefront — there's no retail premises to visit, and the address on file is the operator's registered business address: 7 Oak Lane, Austin, Texas 54422, United States. Every licence sold from this site is issued from that same operating entity; there's no reseller or franchise layer between you and the team building the engine.
VolceMood builds and licenses a real-time voice-changer and audio-effects engine: pitch and formant shifting on two independent dials, a noise gate, and a hotkey soundboard, running as a desktop app, a mobile companion app, and a metered REST API. It sits between a microphone and whatever the customer talks through — Discord, OBS, Zoom, a game's own voice chat — and processes the signal on the device it's installed on, which is checkable by anyone: disconnect the network and the desktop and mobile engine keeps working, because nothing about the voice processing itself leaves the machine. The one exception, stated on the plans page rather than buried, is the Developer API, which is a network call because that's what an API is.
Email [email protected] or call +1 (669) 555-8229. Tickets are answered Monday to Friday, 9:00 to 18:00 Central — two of the nineteen people on the team work support full time, and a licence or refund question goes to a person, not a bot, inside that window. Outside it, a message sent through the inquiry form or the chat widget waits until the next business morning.
Read the engine explanation, then the tiers.
The latency-first stance above is a summary. The full breakdown of the pitch/formant model, the noise gate and platform routing lives on its own page, and the four licence tiers that come out of it are priced by inquiry on Plans.