The short answer
A voice changer takes your microphone signal, transforms it, and sends the transformed version onward to your game, your call, or your stream. A real-time voice changer does this live, while you talk, with only a small delay, so a conversation still feels like a conversation. That is the kind most people mean: you speak normally and the other side hears a deeper voice, a higher one, a robot, or a completely different person.
Under the hood there are two very different ways to change a voice, and knowing which is which tells you what to expect from quality, latency, and control. One reshapes the sound you already have. The other replaces it with a modeled voice.
Family one: DSP effects
The first family is built from DSP effects (DSP is short for digital signal processing), the same kind of audio effects a studio uses, chained together and run live. Nothing here tries to imitate a specific person. Instead, each effect nudges a measurable property of your voice:
- A pitch shifter moves your whole voice up or down, so you sound higher or deeper.
- A formant shifter changes the resonance of your vocal tract, which is what makes a voice read as larger or smaller, independent of pitch.
- Character effects like a ring modulator, a radio or megaphone filter, or a bitcrusher add robotic, broadcast, or gritty textures.
Because these are pure math on the audio, they are fast and predictable. You can stack several of them, tweak the knobs, and hear the result instantly. The tradeoff is that DSP voices sound like your voice processed. That is often exactly what you want for a monster, an alien, or a deeper broadcast tone, but it will not turn you into a specific other person.
A vocoder does not change your voice
One DSP effect breaks that pattern. A vocoder never reshapes your voice at all. It listens to it, and makes a synthesizer say your words instead. It splits your voice into slices of the spectrum, in Patchd from 150 Hz up to 7.5 kHz, and follows how loud each slice is a few milliseconds at a time. That moving pattern of loud and quiet slices is what tells an “ee” from an “oo”. The vocoder splits the synth into the same slices and turns each one up and down to match.
Each slice only ever reports a level, never a note, so your pitch does not make it through. On the take in the video, the voice moved between 127 and 151 Hz while the vocoder held 110.1, the synth's note, the whole time. Set the synth to 220 and the output reads 220.2. That is the robot: your words, on the machine's pitch.
The same split is why a plain vocoder struggles with an s. An s is hiss spread high across the spectrum, and a synth holding one note has no hiss in it to shape. Patchd's Vocoder node answers that two ways: Sibilance lets your real voice through above 4.5 kHz, the one part of it the vocoder keeps, and Noise mixes hiss into the synth so the synth can make the s itself. The synth is built into the node, which is why its note sits on a Carrier control, and Persona's Droid voice is this same vocoder at 16 bands with a chord on the synth.
Family two: AI voice conversion
The second family is AI voice conversion. Here a trained model listens to what you say and re-synthesizes it in a target voice, keeping your words, timing, and inflection but replacing the tone and character of your voice with a different persona. This is how you go from your own voice to a wholly distinct character while still sounding natural and expressive.
The tradeoff is latency and warm-up. A model has to run every moment of audio through a neural network, which takes real compute, so AI voices add more delay than a simple filter and need a moment to spin up the first time you select them. Once running, though, a good implementation holds steady and switches cleanly.
Be honest about latency
Latency is the single most misunderstood thing about voice changers, so here is the plain version. Latency just means the delay between when you speak and when the changed voice comes out. DSP voices add little delay, from none at all for simple effects to tens of milliseconds for pitch-based voices, so you can hear yourself live and it feels immediate. AI voices are different: expect about 245 ms of added delay, plus a few seconds of first-time warm-up while the model loads. After that, switching between AI voices is near-instant.
For most streaming, gaming, and content work that 245 ms is a non-issue, because your voice and your video are lined up together on the way out. It only matters in tight two-way conversation, where the small extra lag is the price of a fully modeled voice. Any tool that promises a dramatic AI voice with zero added delay is glossing over how the technology works.
Where a voice changer lives in your setup
A voice changer is only useful if the right apps actually hear it. On Windows the clean way to do this is through a virtual audio mixer: your mic goes into the mixer, the voice changer processes it, and the result is sent to a virtual output that other apps can pick up as a microphone. If you want to see how it all connects end to end, the voice changer for streaming setup walks through it step by step.
It also helps to feed the changer a clean signal. Background hum and keyboard noise get transformed right along with your voice, so cleaning the input first, for example with noise suppression on your mic, makes any voice changer sound noticeably better.
How Persona does it
Persona is Patchd's voice changer, and it uses both families deliberately. Instant DSP voices are built from real-time effects, the same kind of processing as the mixer's effects rack, so they add far less delay than an AI voice: little to nothing for simple effects, and tens of milliseconds for pitch-based voices. AI voices layer on modeled conversion for the times you want to become a distinctly different character, with the warm-up and delay described above.
Persona applies to any input in the Patchd mixer, and it runs as the final stage of that channel's rack, so cleanup nodes like noise suppression, the noise gate (which mutes the mic when you are not talking), and EQ have already cleaned things up before the voice is applied. Persona also has a face of its own: one click turns the mixer window into a dedicated voice changer for when that is all you want on screen. Switching is hotkey-first (with Stream Deck and Touch Portal support too), so you can flip voices mid-sentence without touching a window.
The free tier includes 6 rotating taster voices so you can try real DSP voices on your own mic, not a demo clip. Persona ships 111 voices, and the Studio tier unlocks the full library, including the AI voices. The lineup is still growing, so treat that number as where things stand today, not a final list. If you just want to hear how a live voice changer sounds right now, the in-browser voice changer is a quick way to get the feel before Patchd ships.
