SpeechLens Download for Mac

Speech enhancement Apple Silicon

Clearer speech.
Right on your Mac.

Enhance audio and video locally with NVIDIA’s RE-USE model. Your audio stays on your Mac.

Download for Mac View on GitHub Watch the demo

Apple Silicon · macOS 14+ · Signed and notarized for macOS

Model weights are licensed for non-commercial research and educational use. Read the model terms

A real before-and-after

Hear the difference

Sample credits

CitySpeechMix · 10 sec · 44.1 kHz · Mono

0:00 / 0:10

Switch versions at the same point in the recording.

Audio stays local

Your media is processed on your Mac.

Native sample rate

Preserves the input sample rate and frame count.

Stereo stays stereo

Preserves channel layout without downmixing.

SpeechLens menu-bar app showing a completed enhancement and original/enhanced playback controls The SpeechLens menu-bar app

A focused Mac utility

A small app.
A straightforward workflow.

  1. 1

    Drop an audio or video file

    Work with common formats, including audio and video.

  2. 2

    Enhance speech on your Mac

    Run RE-USE locally with Swift, MLX and Metal.

  3. 3

    Compare and reveal the result

    Listen to the original and enhanced audio, then find the output in Finder.

On first launch, review the model terms and download the model.

Installation steps

See it in action

From file to clearer speech.

A quick walkthrough of SpeechLens.

The recording shows first-launch setup, adding a file, enhancing speech locally, and comparing the result.

Measured on Apple Silicon

Built with Swift, MLX and Metal.

A native Mac app and CLI, with published performance measurements and source code.

2.3×–6.6× real-time on tested M1 Max inputs

Read the benchmark methodology

Release-mode streaming measurements. Enhancement timing excludes model loading and file I/O; results vary by input and Mac.

Source code CLI guide

Good to know

A few practical details.

What does SpeechLens need?

SpeechLens runs on Apple Silicon with macOS 14 or newer. On first launch, the app asks you to review the model terms and downloads the separate model weights.

Can I use the model commercially?

The SpeechLens source code is Apache-2.0. The separately licensed NVIDIA RE-USE weights are limited to non-commercial research and educational use. Review the upstream model terms and the SpeechLens artifact notes before using the weights.

Does my audio leave my Mac?

Audio and video media are processed locally and are not uploaded. The app has separate, consent-based reliability and performance diagnostics; they do not include audio, filenames or paths. Read the privacy details.

Can I use it from scripts?

Yes. SpeechLens also provides a command-line interface for scripted enhancement. See the CLI instructions in the repository.

Listening sample credits

CitySpeechMix urban noise. Modan Tailleur, Mathieu Lagrange, Pierre Aumond and Vincent Tourre (2025), CitySpeechMix: A Simulated Dataset of Speech and Urban Sound Mixtures from LibriSpeech and SONYC-UST. DOI 10.5281/zenodo.15405950. CC BY 4.0. The enhanced sample was processed with SpeechLens; both listening files were encoded to MP3.

Edinburgh noisy speech. Cassia Valentini-Botinhao (2017), Noisy speech database for training speech enhancement algorithms and TTS models, 2016 [sound], University of Edinburgh DataShare. DOI 10.7488/ds/2117. CC BY 4.0. The enhanced sample was processed with SpeechLens; both listening files were encoded to MP3.