25launches

The first launch of Voice Isolator.

Voice Isolator page

Voice Isolator

Launching today

Open-source model that cleans up a voice in real time

Hunted by IlyasHunter
Visit inworld.ai (opens in a new tab)

Voice Isolator is Inworld's first open-source model. It removes background noise, music and other chatter from speech at 48 kHz while keeping the speaker's timbre, level, breaths and laughs. Two checkpoints: an offline model (64M parameters) for recordings, and a streaming one (58M) for live audio, with 63 ms of algorithmic latency at 20 ms look-ahead. Installs with pip; CLI and Python API on CUDA, Apple silicon or CPU. Code and weights are Apache-2.0. Denoising only for now, no restoration.

Meet the team

Hunted by Ilyas. The makers are not on 25launches yet.

H (Hunter)
  • Voice Isolator screenshot 1
  • Voice Isolator screenshot 2
  • Voice Isolator screenshot 3

Discussion 1

Ask questions, give feedback. Comments don't count in the ranking.

Sign in to join the discussion.

  • IlyasHunter Pinned

    Clean input audio is the first thing a voice agent needs, and the good denoisers are mostly paid APIs. Inworld released both the code and the weights under Apache-2.0, with a streaming checkpoint for live calls. Two things to know: 63 ms is the model's algorithmic latency, not end-to-end time on your hardware, and the benchmark where it ranks first is Inworld's own private set.