Kowane

What it is

Kowane is a Max for Live Audio Effect that turns a voice into a bundle of sine waves: sine-wave speech. Kowane MIDI is a companion MIDI Effect that turns Kowane’s analysis into MIDI notes, so an instrument can follow the voice’s changing sound. Use Kowane on its own, or connect one or more Kowane MIDI devices through their from menus.

Requirements

• Ableton Live 12 Suite, or Live with Max for Live.
• Max 9.
• Tested with Live 12.4 / Max 9.1.4 on macOS.
• Windows: not tested yet.

Install

Unzip Kowane_1.1.zip. The Kowane 1.1 folder contains:

Keep the folder somewhere convenient, such as your Live User Library. Drag Kowane.amxd onto an audio track, or after an instrument on a MIDI track. Drag Kowane MIDI.amxd onto a MIDI track, before its instrument. You can also add the folder to Live’s browser and load the devices from there.

The JavaScript is embedded in the devices; no separate script installation is needed. The device filenames stay the same across this update, so their names in Live do not acquire a version number.

Quick start

1. Put Kowane.amxd on a track playing a voice recording, or a track receiving a microphone. For a microphone, enable the track’s input monitoring so audio reaches Kowane.
2. Play or speak. Kowane’s default settings produce four sine waves, snapped to C major.
3. Create a separate MIDI track and put Kowane MIDI.amxd on it.
4. In Kowane MIDI’s from menu, choose the track containing Kowane.
5. Put an instrument, such as a piano, after Kowane MIDI. Play the voice again to hear the instrument follow it. The default spectrum mode and retrig on work well with decaying sounds.

To hear only the MIDI instrument, set Kowane to sines off. With the default mix of 100%, Kowane’s audio output becomes silent while analysis continues. To include some original voice, lower mix.

Kowane controls

The two columns beside the display separate analysis and time on the left (thresh, voices, step, timing, width, envelope) from pitch and sound on the right (snap, depth, shift, glide, gain, out, mix). Numeric controls support Live’s usual dragging and typed entry. Hover over a control to read its description in Live’s Info View.

Left column

Step durations are approximate: their exact duration depends on the audio sample rate.

Right column

Snap blends pitches on a logarithmic scale: 50% is halfway in musical pitch between the original and the nearest scale note. Kowane’s pitch and audio-output controls do not transpose or mute the analysis sent to Kowane MIDI; set its own scale and shift separately.

Bottom menus

Both devices offer the same scale menu, grouped as follows. These are the labels shown in the menu.

Envelope editor

Click envelope to open or close the editor. When enabled, each step starts a plucked bundle of sines at fixed pitches. Each bundle follows an attack–decay–sustain–release shape and ends after its chosen length. Version 1.1 uses this single pluck behavior; there is no gate/pluck mode selector.

Extending a stage can shorten its neighbours so the shape fits the bundle length. The remaining time between decay and release holds the sustain level. For click control, the audio engine retains a minimum 1 ms attack and 3 ms release, including when the editor shows 0%. Envelope values are saved with the device and are available as Live parameters for automation.

Kowane MIDI controls

Kowane MIDI receives analysis from Kowane and generates notes for the instrument after it, or for an external MIDI destination. Its envelope mode describes the voice’s spectral envelope; it is separate from Kowane’s plucked amplitude envelope.

Choosing a mode

• formant converts each available sine/formant frequency to a note and keeps the strongest candidates up to notes. Kowane’s voices limits the available candidates. Out-of-range notes are folded by octaves into low–high when the range spans at least 12 note positions; for narrower ranges, out-of-range notes are omitted.
• spectrum finds peaks in an 88-key spectrum, including pitch and harmonics, then plays the strongest candidates. This is the starting point for making a piano speak. Candidates outside low–high are omitted.
• envelope maps the smooth spectral envelope of the voice onto the keyboard, selecting its strongest regions rather than individual harmonic peaks. It emphasizes vowel colour and can create clusters. Candidates outside low–high are omitted.

The source selection follows track renaming and reordering during a session. It is saved by track name for reopening a Set; distinct source-track names make that selection unambiguous. Keep one Kowane per source track, because devices on the same track share that track’s analysis destination.

Views

The view menu changes Kowane’s display without changing its sound.

vowel shows the first two formants, F1 and F2, on a chart normalized to the speaker. Higher F1 appears lower in the chart; higher F2 appears further left. White dots show the recent path, the orange dot is the current position, and the nearest vowel label brightens. The labels a, i, u, e, and o correspond to the five Japanese vowels. Normalization adapts while speech arrives, so the position can shift as it learns the speaker.

trace shows the sine frequencies over time, with higher frequencies higher on the display and recent values at the right. White trails distinguish the voices, with stronger emphasis on lower voices; orange markers show the current values.

lips, the default, draws a mouth with pink dots. Higher F1 opens the jaw; higher F2 widens the lips. Input level also affects opening and brightness, and the mouth closes during silence. A nearby vowel label appears at the lower left. This is a visualization derived from formants, not a measurement of the speaker’s actual mouth.

Kowane MIDI has a note display rather than a view menu. Horizontal position indicates pitch within low–high; dot size indicates velocity. Note strikes rise and fade, while orange dots along the bottom show notes currently held.

Tips

• Try drums and instruments. Set Kowane’s source to full to retain the low end without speech pre-emphasis. Return to voice for speech.
• Drive several instruments from one voice. Put Kowane MIDI before each instrument and choose the same source track in every from menu. Give each instance its own mode, note count, range, scale, and rate.
• Reduce CPU use. Lower Kowane’s voices. When its envelope is enabled, lower layers to reduce overlapping bundles.
• Recover the stepped pitch movement of sinewavespeech.com. Set glide to 0 ms and use timing classic with the envelope off. no scale, depth 0, shift 0, and width 0 remove Kowane’s additional pitch and stereo changes.
• Use larger steps for slower changes. A larger step collects a longer slice of input before analysis can finish. classic then uses the previous step’s pitches and ramps levels across another step, so increasing step size also increases audible delay.

Troubleshooting / FAQ

Kowane is silent.

Check that audio reaches the device and that its track is audible. Lower thresh if the voice is being treated as silence. Check sines: sines off mutes the synthesized sound but leaves the dry level at the current mix setting. At mix 100%, this is silent; at mix 0%, you hear only the original input. Also check out and the track level.

No MIDI arrives, or the source does not appear in from.

Put Kowane on the separate source track and confirm that it is receiving audio. Put Kowane MIDI before an instrument on the receiving MIDI track, or provide that track with a valid external MIDI output. Select the source explicitly in Kowane MIDI’s from menu. The source audio track does not need a MIDI output: analysis is passed between the devices. The menu looks for a device name containing Kowane without MIDI, so retain that name if you rename devices. For missing notes with a connected source, check Kowane MIDI’s gate, low, and high; in formant mode also check Kowane’s thresh.

What does no signal mean?

A source is selected, but Kowane MIDI has not received the source’s status message for about 1.5 seconds. Check the source device and Live’s audio engine. This reports missing communication, not simply an input level below the gate. If analysis stops, a separate timeout releases generated notes after roughly 250 ms, checked every 100 ms.

There is too much delay.

Reduce Kowane’s step and select timing tight. In Kowane MIDI, a low rate also spaces note updates further apart. Live’s audio buffering and the receiving instrument can add their own delay.

Snap seems inactive.

Select a scale other than no scale. Snap is dimmed and has no effect with no scale; its value is retained. Set it to 100% to hear the strongest pull toward the selected notes.

Glide is dimmed.

Kowane’s envelope is on. Plucked bundles keep fixed pitches, so glide is unavailable. Turn the envelope off to use glide.

Version history

1.0 — 2026-10-27

Initial release of Kowane and Kowane MIDI.

1.1 — 2026-10-27

• Envelope now uses a single pluck behavior; the former gate mode has been removed. Choose length in steps (1–16) or milliseconds (20–2000), with 1–8 overlapping layers. Replacing the oldest bundle uses a 5 ms fade to remove clicks, and the editor now allows attack to be set to 0%.
• Added width (0–100%) to spread the sines in stereo.
• Changed snap from the former 0–3 stages to a continuous 0–100% pull toward the selected scale. Snap is dimmed when no scale is selected.
• Added glide (0–100 ms, default 10 ms) to smooth pitch changes. Set it to 0 ms for the original stepped sound.
• When updating Sets made with 1.0, revisit snap: values from the former 0–3 range are small amounts in the new percentage range. Reset snap to the amount you want.

Credits / license

Kowane — Cotofu 2026.

Kowane’s core sine-wave speech analysis and synthesis build on sine_wave_speech by Václav Volhejn (MIT License); portions of that code were adapted for Kowane. Everything around it — the Max for Live devices, scale snapping, glide, width, the envelope, the views and Kowane MIDI — is Kowane’s own. The full MIT license text is included in the zip as LICENSE-sine_wave_speech.txt and is available at cotofu.com/licenses/.