BIRDSPEECH

0:00
drop file, record mic, or pick sample
0:00
FAQ
What does it do?

This site turns human voice into birdsong. It's inspired by windytan's Blackbird experiment

How does it work?

A voice is a stack of harmonics over a fundamental pitch. The converter shifts that fundamental down to near 0 Hz, low-pass filters it to isolate it, and then uses FM demodulation to track its instantaneous frequency. AM demodulation on the same signal gives the amplitude envelope. The original harmonics are thrown away. The output is a freshly synthesized tone six times higher, shaped by the tracked pitch and envelope.

That is why it reproduces melody, rhythm and inflection so well but not words: only F0 and loudness survive, not the formants that make phonemes. Windytan's original ran on a Perl/SoX/csdr pipeline. This version ports the same DSP to an AudioWorklet, with extra pitch, timbre and trill controls.

Settings tips?

"Voice range" is the most important value, change it to fit your low and high vocals. The "auto" button will try to guess it from the media file.

Hitting "Randomize" is very fun. You can checkbox which values you want randomized or kept.

Voice tips?

The best input is exaggerated spoken word, poetry, or slow song acapellas. Try and make it rhythmic, like a bird song!

The conversion works best with clean vocals and no background noise. But you can try adjusting the Equalizer curve to try and focus audio range.

What birds can it do?

It can do birds with a clean voice: small singing birds, but also some bigger ones like pigeon, cuckoo, owl. It cannot do ducks' croak or crows' caw, they would need a different approach.

Mixto randomize ↴  
100%Blend input voice (0%) with birdsong (100%)
15%Everybody knows what reverb is
Bird presetsclick to load
Bird voiceto randomize ↴  
6.0×How many times higher the bird sings than the tracked voice. Slide up for tiny birds. Down for bigger deeper birds.
1How many overtones are layered above the fundamental.
sineThe overtone pattern the Timbre control counts up.
0%Adds a fast pitch wobble
14 HzHow fast the trill wobbles
Voice trackingto randomize ↴  
220 HzThe part of the voice the tracker listens to. Change for low or high voices.
140 HzHow wide a slice of the voice is tracked. Wider catches more overtones and sounds raspier, narrower stays cleaner.
0%Raise for noisy recordings.
70 HzHow closely the bird follows quick pitch changes. Low is smooth, high is wobbly.
70 HzHow quickly the bird's loudness follows the voice. Low is smooth, high is precise.