Inworld AI · Voice Isolation Model

Inworld Voice Isolator v1

Inworld Voice Isolator v1 is a 48 kHz speech denoising and voice isolation model released under Apache-2.0, with code and weights on Hugging Face and GitHub. It ships as two checkpoints sharing one architecture: an offline model (64M parameters) that sees whole 8 s chunks, and a causal realtime model (58M parameters) with 20 ms of look-ahead and 63 ms algorithmic latency. The page gives no benchmark scores, noting that the VIBE results rely on an internal evaluation set that is not distributed, and says restoration is planned for version 2.

Open the source ↗

Similar Audio & Speech releases

  • ACE Music v1

    ACE Music · Vocal music generation · Week 47, 2026

  • Haru 4B

    higurashi · Full-duplex speech model · Week 47, 2026

  • LiquidCodec

    Liquid AI · Streaming Audio Codec · Week 47, 2026

  • MOSS-Live

    模思智能 (MOSS) · Full-duplex voice agent · Week 46, 2026

  • Demucs

    Meta AI Research (facebookresearch) · Audio source separation · Week 46, 2026

  • NVIDIA Nemotron 3.5 Transcribe 30B-A4B

    NVIDIA · Open multilingual ASR · Week 45, 2026

More from Inworld AI

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.