nymic

Real-Time · On-Device · iOS

Your Voice,
Any Voice.

nymic transforms your voice in real time. kNN-VC AI voice conversion makes you sound like a completely different person — plus DSP effects for Robot, Alien, Deep Bass, and more — all on-device. No cloud, no latency, no compromise.

Download for iPhone Contact Support

0

Cloud Servers

30

AI Voice Profiles

100%

On-Device

How It Works

Built Different. Literally.

nymic captures your mic, processes through a custom C++/Metal DSP engine and ONNX Runtime ML pipeline. ~500–600ms end-to-end for AI voice conversion.

Voice Library — Bank-Swappable

Each voice is a 4–60 MB feature bank. The WavLM and HiFiGAN models (402 MB total universal backbone) are shared — swap voices in under a millisecond without reloading the ML backbone. One model, infinite voices. Library grows with curated public-domain narrators and on-device custom clones.

Clone Your Voice On-Device

Record 15+ seconds of speech or import an audio file (WAV, M4A, MP3). WavLM extracts features locally via ONNX Runtime, builds a feature bank, and saves it — a new selectable voice in under a minute. Audio never leaves your iPhone.

AI Voice Conversion

kNN-VC powered by WavLM-Large (339 MB) + HiFiGAN (63 MB) via ONNX Runtime. Sound like a completely different person — genuine timbral and phonetic transformation, not just pitch shifting. ~500–600ms end-to-end latency on recent iPhones; the live monitor is a best-effort preview and the saved export is a full-quality offline re-render produced on stop.

Record & Export

Tap to capture your transformed voice as high-quality WAV (16 kHz, 32-bit). Export via AirDrop, Messages, or Files. Biometric lock keeps recordings private behind Face ID / Touch ID.

Audio Pipeline

From Mic to Speaker

Microphone

24 kHz float32

→

Ring Buffer

1024-sample frames

→

kNN-VC ML

WavLM + kNN + HiFiGAN

→

Overlap-Add

WSOLA overlap-add

→

Speaker Out

Real-time

DSP profiles included:

Original Deep Bass Ethereal Broadcast Neural (ANE) Robot Alien Telephone Megaphone Phaser Underwater Whisper Fuzz Chipmunk Titan Metal Gain Formant Shift Dynamic Saturation Transient Morph

Hear It in Action

Before & After — The Real Sound of nymic

Play the raw voice, then toggle to hear the transformed output. Real processed audio, not simulations.

Male → Female

AI Voice Conversion · Cross-Gender

Switch to Transformed

Male input → female output via kNN-VC

Male → Male

AI Voice Conversion · Same-Gender

Switch to Transformed

Male input → male output via kNN-VC

Designed for iPhone

Swipe Through the Experience

Find Your Voice

Which Voice Are You?

Answer three quick questions to discover the nymic voice profile that fits your style.

Try Again

Question 1 of 3

Use Cases

One App, Infinite Characters

The Roleplayer

DSP profiles — Robot, Alien, Titan, Whisper — built for tabletop RPGs and immersive gaming. Speak as any character without breaking flow.

The Creator

Record transformed voice as WAV. Export to your DAW, video editor, or podcast. AI voice conversion for identity-level transformation — no studio required.

The Privacy Advocate

All processing on-device. Zero cloud. Zero analytics. No account required. Your voice never leaves your iPhone — guaranteed by architecture, not policy.

Why nymic

Not All Voice Changers Are Equal

Most voice changers upload your audio to cloud servers, apply simple pitch shifts, and call it a day. nymic is built different from the silicon up.

Capability	Other Voice Changers	nymic
Audio Processing	Cloud servers — your voice leaves your device	100% On-Device
Voice Effects	Canned presets, single pitch shift	Custom C++/Metal DSP profiles
AI Voice Conversion	None or cloud-only	kNN-VC: WavLM + HiFiGAN on-device
Voice Cloning	Cloud upload required	On-device, audio never leaves
Offline Use	Requires internet	Works fully offline
Privacy Model	Analytics, tracking, data collection	Zero analytics, zero tracking, no account
DSP Engine	Basic audio framework presets	Custom C++ biquads, vocoder, ring mod, phaser, grain pitch
Export Quality	Compressed, lossy	WAV at 16 kHz, 32-bit

Technical Specs

Engineering That Speaks for Itself

Specification	Detail
DSP Engine	Custom C++/Metal pipeline with biquad filters, delay lines, ring modulation, all-pass phasers, 4-band vocoder, grain-based pitch shifting, tanh saturation, and Metal GPU compute
ML Voice Conversion	kNN-VC pipeline: WavLM-Large (339 MB, layer 6 feature extractor) → cosine kNN matching with top-4 uniform averaging and Viterbi transition penalty for phoneme-contiguous retrieval → prematch HiFiGAN neural vocoder (63 MB). Per-hop inference ~265 ms on A19 Pro; end-to-end latency (including 1 s analysis window and buffering) ~500–600ms. Live monitor auto-engages INT8 HiFiGAN on older devices or under thermal load; saved export uses full FP32 quality offline re-render.
Audio Pipeline	AVAudioEngine microphone tap at 24 kHz mono float32 → ring buffer architecture → DSP or ML processing path → AVAudioSourceNode output. 1024-sample buffer size, overlap-add streaming with WSOLA phase alignment for artifact-free continuous conversion
Hardware Acceleration	CPU (4-thread ONNX Runtime intra-op parallelization), Metal GPU (gain pipeline), Apple Neural Engine / ANE (Core ML profile 5 — Neural identity reconstruction)
Voice Library Architecture	Multi-bank kNN-VC — 402 MB universal backbone (WavLM 339 MB + HiFiGAN 63 MB) shared across all voices, plus 4–60 MB per-voice feature bank. Bank-swappable in under 1 millisecond without reloading backbone models. Includes bundled Default Narrator (LibriSpeech 211 corpus, shanda_w, 15,000-row bank) and supports on-device custom voice cloning
Device Requirements	iOS 18.0 or later · iPhone XS (Apple A12 Bionic chip) or newer · iPhone 14 Pro or newer recommended for optimal AI voice conversion performance; AI conversion hardware-gated on older/low-memory devices
Export Formats	WAV at 16 kHz 32-bit. Share via AirDrop, Messages, iOS Files app, or system share sheet
Background Audio	Supported via UIBackgroundModes: audio entitlement — voice processing continues when the app is backgrounded
Storage Footprint	Approximately 407 MB for the kNN-VC pipeline (WavLM-Large 339 MB + HiFiGAN 63 MB + feature bank ~5 MB). Well within Apple's 4 GB IPA size limit. Additional local storage used only for user-saved WAV recordings

Privacy & Security

Your Voice Never Leaves Your iPhone

Every bit of audio processing — DSP and ML alike — happens on-device. Zero cloud. Zero analytics. Zero exceptions. Privacy enforced by architecture, not policy.

No Cloud Processing

All DSP and ML inference runs on your iPhone's CPU, GPU, and Neural Engine. Your microphone audio is never uploaded, streamed, or stored on any server. There is no server.

Zero Data Collection

No analytics. No telemetry. No third-party tracking SDKs. No account required. Install and use — we don't know who you are by design.

Biometric Protection

Lock recordings behind Face ID or Touch ID. All data stored in the iOS application sandbox, inaccessible to other apps or system services.

No Internet Required

All core functionality works fully offline. ML models are bundled in the app. Internet is only needed for the initial App Store download and optional model updates.

Technical Privacy Architecture

nymic's privacy guarantees are enforced by architecture, not policy:

• In-Memory Processing — Microphone audio is processed through a ring buffer and immediately discarded after playback. No audio is persisted unless you explicitly tap the record button.
• iOS Sandbox Enforcement — All model inference and audio processing runs within iOS's per-app sandbox. No access to other apps' data or filesystem outside its container.
• Network Isolation — nymic does not declare network entitlements for core functionality. ONNX Runtime, C++ DSP, and Metal GPU all operate without network access.
• In-Memory Feature Extraction — Custom voice cloning extracts WavLM features entirely on-device. Feature banks are stored locally in the app sandbox. No audio is transmitted during cloning.
• Complete PrivacyInfo.xcprivacy — Full Apple-required privacy manifest declaring all API usage: microphone, file timestamps, system boot time, disk space, UserDefaults. No tracking domains declared.

Your Rights

• No account required. No sign-up, no login, no user database. Install and use immediately.
• Your recordings, your control. WAV files saved locally. Delete anytime. Export via system share sheet. You decide what to share.
• Deletion is instant. Deleting the app removes all local data. No backups, no cloud copies, no retained data — because no server exists.
• No tracking, ever. We do not fingerprint your device, profile your usage, or build advertising segments.

Questions? makersportal.com/contact. Because all processing is local, there is no user data for us to provide, modify, or delete — everything lives on your device under your full control.

Questions

Frequently Asked

What is nymic?

nymic is a real-time voice changer for iPhone. Speak into your mic, pick a DSP profile (Robot, Deep Bass, Chipmunk, etc.) or enable AI voice conversion to sound like a different person. All processing happens on-device — no cloud, ~500–600ms for AI voice conversion on recent iPhones.

What's the difference between DSP and AI voice conversion?

DSP profiles are traditional audio effects (Robot, Deep Bass, Chipmunk, etc.) that modify your voice using biquad filters, ring modulation, pitch shifting, vocoding, and other signal processing techniques.

AI voice conversion uses machine learning (WavLM-Large + kNN matching + HiFiGAN) to make you sound like a completely different person. It analyzes your phonetic content, matches it against a target speaker bank, and synthesizes new audio through a neural vocoder. ~500–600ms end-to-end latency on recent iPhones. This is actual voice transformation, not just an effect.

Does nymic require an internet connection?

No. All voice processing — DSP effects and AI voice conversion — runs entirely on-device. The ONNX Runtime ML models and C++/Metal DSP engine do not require a network connection. Internet is only needed for the initial App Store download. Use nymic on an airplane, in a tunnel, or anywhere without a signal.

Is my voice recorded or sent anywhere?

No. Your microphone audio never leaves your iPhone. There is no cloud server, no backend, and no analytics pipeline. Audio is processed in memory through a ring buffer and discarded immediately after playback — it is never saved to disk unless you explicitly tap the record button. Even then, recordings are stored locally in the iOS sandbox and are never uploaded.

Which iPhones does nymic support?

Requires iOS 18.0+ and an iPhone XS or newer (Apple A12 Bionic chip). For the best AI voice conversion experience, iPhone 14 Pro or newer is recommended. AI voice conversion is hardware-gated on older/low-memory devices — DSP effects are available on all supported devices. The app uses Metal GPU compute, the Apple Neural Engine (ANE), and ONNX Runtime with multi-threaded CPU parallelization for efficient on-device inference.

What makes nymic different from other voice changers?

Three things set nymic apart:

1. 100% on-device processing. Most voice changers upload your audio to cloud servers for processing. nymic runs everything locally — C++/Metal DSP and ONNX Runtime ML — so your voice never leaves your iPhone. This means lower latency, offline capability, and genuine privacy.

2. Real DSP, not presets. nymic's voice profiles are hand-tuned effects running on a custom C++ audio engine with biquad filters, delay lines, ring modulation, all-pass phasers, a 4-band vocoder, and granular pitch shifting. Most voice changer apps apply a single pitch shift or a canned preset.

3. Actual AI voice conversion. Beyond DSP effects, nymic includes a full kNN-VC machine learning pipeline (WavLM-Large feature extraction + cosine kNN matching + HiFiGAN neural vocoder) that transforms your voice to sound like a different person — genuinely different in timbre and phonetic quality, not just pitch-shifted.

Can I record and export my transformed voice?

Yes. Tap the record button to capture live sessions as high-quality WAV files at the device's native hardware sample rate (16 kHz, 32-bit). Export via AirDrop, Messages, Files, or any compatible app. No time limits. Recordings can be locked behind Face ID / Touch ID via the biometric lock setting.

Does nymic work with Bluetooth headphones?

Yes. nymic works with built-in mics, wired headsets, and Bluetooth devices including AirPods. System-level acoustic echo cancellation is enabled to prevent speaker-to-mic feedback. For best AI conversion quality, use the built-in mic or a wired headset — Bluetooth HFP can be bandlimited, which affects conversion accuracy.

How much storage space does nymic need?

The app binary plus bundled ML models total approximately 407 MB for the kNN-VC pipeline (WavLM-Large 339 MB + HiFiGAN 63 MB + feature bank ~5 MB). Additional storage is used locally only for WAV recordings you choose to save — these can be managed and deleted from the in-app history view. Model weights are loaded on-demand and shared between inference sessions to minimize memory pressure.

Can I use nymic with Discord, Zoom, or phone calls?

nymic is a standalone voice changer app. It transforms your voice in real time through the device speaker or headphones.

Can nymic make me sound like a specific person?

nymic's kNN-VC is a zero-shot voice conversion system — it transforms your voice toward a target speaker's timbre by matching your speech features against a pre-built feature bank of that speaker's voice. The underlying architecture supports any target speaker given sufficient high-quality reference audio. The on-device voice cloning feature lets you build a feature bank from any audio source (WAV, M4A, MP3) entirely on-device. DSP profiles let you dramatically alter your voice instantly without training data.

nymic

© 2026 Maker Portal LLC — Real-time on-device voice changer for iPhone

Maker Portal Privacy Policy Support