What exactly does the radio preset do to my voice?

Two things, in order. The band-pass is centered on 1,500 Hz with a Q of 1, so it passes about 930 Hz to 2,430 Hz at no more than 3 dB down and cuts further the further a sound is from that band. That removes the chest and the air of a voice, the parts a small speaker cannot reproduce. The distortion then applies tanh(6x) / tanh(6): quiet passages come out about six times louder, and the loudest ones are pressed up against full scale without wrapping, which is the overdriven, crackly edge. Ring mod and delay are off, the mix is 100% wet and the rate is 1, so the clip keeps its length and pitch.

How do I make it sound more like a walkie-talkie or a phone?

Narrow the band. Raise Band-pass Q from 1 toward 3 or 4 and the passband shrinks around 1,500 Hz, which sounds more like a cheap handset. Raise the drive for more crunch, lower it toward 1 for less crunch, or switch Distortion off for a clean but thin speaker. Any change shows the preset as Custom, and the Radio chip puts the original values back.

Why does the result sound quieter or louder than my recording?

The band-pass throws away energy outside its band, so a full-range voice loses level there, and the drive then multiplies what is left by up to six. Which wins depends on the recording. The clipping line after each render counts samples past full scale and gives the peak in dBFS. Turn on Normalize if you want the loudest sample set to -1 dBFS.

Radio Voice Effect

This page opens the voice changer on its Radio chain, the thin, crunchy sound of a voice coming out of a small speaker, which is also what people mean by a walkie-talkie or bad microphone voice. Two stages make it. A band-pass centered on 1,500 Hz with a Q of 1 keeps roughly 930 Hz to 2,430 Hz within 3 dB of full strength and lets the lows and highs fall away, and a tanh distortion with drive 6 lifts quiet sound about six times while rounding off the loud peaks. Record a clip, open a file or try the built-in voice, switch Preview between dry and wet, and download the result as a WAV. Every value stays editable, and the audio never leaves this browser tab.

Open a voice recording

· ·

or drop an audio file anywhere on this panel

Never uploadedNo accountUp to 50 MiBUp to 5 minutes

Preset: Radio
Ring mod: off
Export: no audio yet
WAV: no audio yet

Ready. Open a recording, record a clip or try the built-in voice.

440 Hz sine x 80 Hz sinering modulation, measured
sidebands at 360 Hz and 520 Hz, -6.0 dB and -6.0 dB; 440 Hz below -100 dB
Rate 2 alonememo.wav, 12.500 s
6.250 s = 300,000 samples, every pitch doubled
Wet mix 0any chain, rate 1
the input PCM back, largest difference 0 (tolerance 1e-6)

Effect chain

The voice runs through these in order, top to bottom. Every change renders again straight away. Arrow keys step a field up or down; Enter or leaving the field applies what you typed.

1 Band-passCenterQkeeps a band around the center, unity gain there
2 Ring modFrequencyoff: when on, it multiplies by a sine at 80 Hz
3 DistortionDrivetanh(drive x) / tanh(drive): full scale stays full scale
4 DelayTimeFeedbackechoes kept until they fall below -60 dB, 5 s at most
5 MixWet100% wet, 0% dry
6 RateRatepitch 0.00 semitones, length x1.00: they move together
7 Normalizeoff: the level is left exactly as the chain makes it

Settings JSON

The chain above as plain JSON, to keep, share or load back later. Loading settings changes the chain, never the recording.

{
  "tool": "voice-effects-studio",
  "version": 1,
  "chain": {
    "bandpass": {
      "on": true,
      "centerHz": 1500,
      "q": 1
    },
    "ringMod": {
      "on": false,
      "hz": 80
    },
    "distortion": {
      "on": true,
      "drive": 6
    },
    "delay": {
      "on": false,
      "ms": 12,
      "feedback": 0.5
    },
    "mixPercent": 100,
    "rate": 1,
    "normalize": false
  }
}

Paste settings JSON

Open a recording up to 50 MiB and 5 minutes, record one here, or try the built-in voice, which is synthesised in this tab. Press Play or Space to hear it, and the Preview chip to switch between A, the recording as it came in, and B, the effect chain, at the same moment. Download WAV writes the finished result B plays as 16-bit PCM. Nothing is uploaded, and only the chain settings are remembered in this browser, never the audio.

Accuracy. Deterministic effects, not AI: in the same browser, the same recording and settings always give the same WAV, echoes are kept until they fall below -60 dB (5 seconds at most), and clipping is counted rather than hidden. Changing the rate moves pitch and length together, and there is no neural voice conversion, cloning, accent change, disguise of who is speaking or live microphone routing. Audio up to 50 MiB and 5 minutes is decoded in your browser after a format check and is never uploaded.

Common questions

What exactly does the radio preset do to my voice?: Two things, in order. The band-pass is centered on 1,500 Hz with a Q of 1, so it passes about 930 Hz to 2,430 Hz at no more than 3 dB down and cuts further the further a sound is from that band. That removes the chest and the air of a voice, the parts a small speaker cannot reproduce. The distortion then applies tanh(6x) / tanh(6): quiet passages come out about six times louder, and the loudest ones are pressed up against full scale without wrapping, which is the overdriven, crackly edge. Ring mod and delay are off, the mix is 100% wet and the rate is 1, so the clip keeps its length and pitch.
How do I make it sound more like a walkie-talkie or a phone?: Narrow the band. Raise Band-pass Q from 1 toward 3 or 4 and the passband shrinks around 1,500 Hz, which sounds more like a cheap handset. Raise the drive for more crunch, lower it toward 1 for less crunch, or switch Distortion off for a clean but thin speaker. Any change shows the preset as Custom, and the Radio chip puts the original values back.
Why does the result sound quieter or louder than my recording?: The band-pass throws away energy outside its band, so a full-range voice loses level there, and the drive then multiplies what is left by up to six. Which wins depends on the recording. The clipping line after each render counts samples past full scale and gives the peak in dBFS. Turn on Normalize if you want the loudest sample set to -1 dBFS.

Deterministic effects, not AI: in the same browser, the same recording and settings always give the same WAV, echoes are kept until they fall below -60 dB (5 seconds at most), and clipping is counted rather than hidden. Changing the rate moves pitch and length together, and there is no neural voice conversion, cloning, accent change, disguise of who is speaking or live microphone routing. Audio up to 50 MiB and 5 minutes is decoded in your browser after a format check and is never uploaded.