🎛️ Geeky Kokoro Advanced Voice (2025)

Every TTS spits out the same polite neutral voice unless you do something about it. 🎛️ Geeky Kokoro Advanced Voice (2025) is the "do something about it" step: you feed it the audio output of the pack's 🔊 Geeky Kokoro TTS node (or any AUDIO in ComfyUI) and it comes back out warped, filtered, and generally a lot more interesting. The headline feature is 18 one-click voice profiles - Cinematic, Monster, Robot, Chipmunk, Darth Vader, Radio, Telephone, Whisper, Shout - which is the fastest way I know to make a character voice without touching a pitch knob.

How it works

It's a DSP chain, not a neural voice conversion. Pitch shifting, formant shifting, reverb, echo, distortion, compression, a 3-band EQ, and time-stretching all run on the waveform, plus an effect_blend that crossfades the processed result against the original - so you can dial in "Monster, but mostly still a person." The profile_intensity scales how hard a profile hits. If you want the raw sliders instead of a preset, flip manual_mode (or pick the Custom profile) and the individual controls light up.

One honest caveat, straight from the README: the author labels this node "under construction" and says it "will not function as intended at the moment." The profiles and manual effects are the more settled half; treat them as functional. The newer stuff is the part to be careful with.

The inputs that matter

Required: audio (the TTS output), voice_profile (the dropdown with 18 choices including None and Custom), profile_intensity (default 0.7), effect_blend (default 1.0 - the "how much of the effect survives" dial, and your best friend when something sounds overcooked), and output_volume (-60 to +60 dB; 0 is unity).

Then the experimental half, which you should approach with lowered expectations: guide_audio + enable_guided_morph let you morph the TTS toward a reference audio file - the README's pitch example is making a voice "sing" along to a track. It's genuinely clever (pitch tracking via PYIN, DTW alignment, spectral transfer) and it's also heavy: the README budgets +800MB to 1.5GB of RAM and a 30–50% slowdown for it. The four *_morph_amount sliders (pitch, formant, spectral, amplitude) are all 0–1.

For the manual controls, the ones beginners actually touch: pitch_shift (±12 semitones) for that obvious "chipmunk/deep" effect, formant_shift for the subtler "same pitch, different body size" character change, and reverb_amount/echo_delay when it sounds too dry.

The only output is audio, same AUDIO type - chain it into a save or preview node, or stack it before a VHS filter if you're feeling bold.

Installing it

It ships in the same pack as the TTS node, so there's no separate install. Whatever got you the 🔊 Geeky Kokoro TTS node gets you this one - ComfyUI Manager (search "Geeky Kokoro TTS") or:

cd ComfyUI/custom_nodes
git clone https://github.com/GeekyGhost/ComfyUI-Geeky-Kokoro-TTS
cd ComfyUI-Geeky-Kokoro-TTS
pip install -r requirements.txt

That requirements.txt is where the DSP heavyweights live - librosa, scipy, resampy - so expect a real install, not a blink. Restart after.

Troubleshooting

• Sounds distorted → drop effect_blend to 0.3–0.5 first, then lower output_volume; distortion and compression are the usual culprits if you've pushed them.
• Guided morph does nothing → enable_guided_morph must be on and guide_audio connected; also keep the reference above ~16kHz sample rate per the README.
• Slow / heavy → the morphing path is the expensive one; disable it for quick previews and re-enable for finals.
• Node missing entirely → restart ComfyUI and check the console - it's imported separately from the TTS node, so one can fail while the other loads.

Reach for this when your narration needs a character, not a personality transplant.