Desert Ant Labs

Clear: Studio sound, on-device speech enhancement

Speech enhancementAvailable

On-device speech enhancement that turns a noisy recording into a podcast-ready 48 kHz file: denoised, dereverbed, and loudness-normalized, all on device.

Clear takes noisy mono or stereo audio, a laptop mic, an untreated room, traffic, and returns a clean 48 kHz result. A fine-tuned DeepFilterNet 3 model handles noise reduction and dereverberation, and an optional pass normalizes loudness to platform spec. Two variants: clear-studio, bundled and used by default, cleans down to near silence; clear-natural, downloaded on request, keeps room tone, breath, and lip texture for treated rooms and intentional voiceover.

Clear works on any audio or video file AVFoundation can read; video comes back with the original picture and enhanced audio muxed in. A 5-minute clip runs in about 2.6 seconds on an iPhone 17 Pro, 114x realtime, measured across the full pipeline including mastering and re-encode. Clear is the on-device alternative to cloud tools like Adobe Podcast, Dolby, and Auphonic: audio never leaves the device, with no upload and no per-minute bill. The model leads with dereverb and podcast-ready output, the qualities most denoisers skip.

Swift SDK Kotlin SDK JS SDK Model card

Demo

Performance

Enhances a 5-minute clip in about 2.6 s on an iPhone 17 Pro, 114x realtime, on device.

114x
Realtime, mono
9.2 MB
Core ML model
48 kHz
Output

5-minute mono clip, full pipeline including mastering and re-encode, iOS 27.0 (internal)

DeviceCPU (default)All compute units
iPhone 17 Pro114x109x
iPhone 16 Pro95x83x

Use cases

Podcast and video mastering

Clean a laptop-mic or untreated-room recording and normalize to Apple Podcasts, Spotify, YouTube, or EBU R128 in a single pass.

Dereverberation

Remove room reflections, not just hiss. Dereverb is where most denoisers stop and Clear keeps going.

Fast and local

A 5-minute clip enhances in about 2.6 seconds on an iPhone 17 Pro, with no upload and no per-minute cost.

Voice and meeting capture

Improve the intelligibility of voice notes and recorded calls before transcription or archiving.

What it does

  • Denoise plus dereverb from a fine-tuned DeepFilterNet 3 model.
  • Loudness presets: Apple Podcasts, Spotify, YouTube, EBU R128.
  • Strength and Mastering controls.
  • Works on any audio or video AVFoundation can read; video returns enhanced audio muxed with the original picture.
  • Two variants: clear-studio bundled for a quiet studio cleanup, clear-natural downloaded on request to keep room tone.

Platforms and install

iOS, macOS
Install
// Swift Package Manager
.package(url: "https://github.com/Desert-Ant-Labs/clear-swift", from: "0.1.0")
Example - Swift
import Clear

let clear = try await Clear()
let result = try await clear.enhance(at: inputURL)
// result.outputPath: "in_clear.mp4"
SDK on GitHub
Android
Install
// build.gradle.kts (via JitPack)
implementation("com.github.Desert-Ant-Labs.clear-kotlin:clear:0.1.0")
Example - Kotlin
import ai.desertant.clear.Clear

val clear = Clear.create(context)
val result = clear.enhance(inputWavPath)
// result.outputPath: ".../<stem>_clear.wav"
SDK on GitHub
Web, Node.js
Install
npm i @desert-ant-labs/clear
Example - TypeScript
import { load, decodeToMono, encodeWav } from "@desert-ant-labs/clear";

const clear = await load({ variant: "studio" });
const { audio } = await clear.enhance(await decodeToMono(file));
SDK on GitHub
Add Clear with your AI coding assistant
Prompt
Add Clear from Desert Ant Labs to this project.

On-device speech enhancement that turns a noisy recording into a podcast-ready 48 kHz file: denoised, dereverbed, and loudness-normalized, all on device.

SDKs (use the one that matches this project's platform):

Swift (iOS, macOS)
Repo: https://github.com/Desert-Ant-Labs/clear-swift
// Swift Package Manager
.package(url: "https://github.com/Desert-Ant-Labs/clear-swift", from: "0.1.0")

Kotlin (Android)
Repo: https://github.com/Desert-Ant-Labs/clear-kotlin
// build.gradle.kts (via JitPack)
implementation("com.github.Desert-Ant-Labs.clear-kotlin:clear:0.1.0")

JavaScript / TypeScript (Web, Node.js)
Repo: https://github.com/Desert-Ant-Labs/clear-js
npm i @desert-ant-labs/clear

Reference:
- Model page: https://desertant.com/models/clear/
- Full catalog and other models: https://desertant.com/llms.txt

Add the SDK that matches this project's platform, then follow its README for the exact API and current version. Do not invent API names or method signatures; confirm them against the README.

Specs

Output
48 kHz, denoised and dereverbed
Speed
114x realtime on an iPhone 17 Pro
Model
Fine-tuned DeepFilterNet 3
Standards
LUFS / EBU R128 loudness normalization
Variants
clear-studio bundled, clear-natural on request
Size
9.2 MB Core ML, 24 MB ONNX, per variant
Platforms
iOS 17+, macOS 14+ (Core ML); Android 8+ and the browser (ONNX Runtime)

FAQ

What is Clear?

On-device speech enhancement that turns a noisy recording into a podcast-ready 48 kHz file: denoised, dereverbed, and loudness-normalized, all on device.

Does Clear run on device?

Yes. Clear runs entirely on device: inference happens locally with no server call, so data never leaves the device.

Which platforms does Clear support?

Clear ships as native on-device SDKs for Swift, Kotlin, JavaScript / TypeScript.

How much does Clear cost?

Every model is free up to 100k monthly active devices per SDK. Unlimited inference per user. Contact us for custom licenses.

How accurate or fast is Clear?

Enhances a 5-minute clip in about 2.6 s on an iPhone 17 Pro, 114x realtime, on device.

Resources

Model weights on Hugging Face Full demo on Hugging Face Swift models Kotlin models JS models