# Adapting Speech Separation to Real-World Meetings Using Mixture Invariant Training

arXiv: https://arxiv.org/abs/2110.10739 · HTML: https://mlsanity.com/paper/2110.10739/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 20 Oct 2021.

## Built on by (most cited first, 8 of 8)

- [MixCycle: Unsupervised Speech Separation via Cyclic Mixture Permutation Invariant Training](https://mlsanity.com/paper/2202.03875/) (2022)
- [ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior](https://mlsanity.com/paper/2505.05657/) (2025)
- [Speech separation with large-scale self-supervised learning](https://mlsanity.com/paper/2211.05172/) (2022)
- [UNSSOR: Unsupervised Neural Speech Separation by Leveraging Over-determined Training Mixtures](https://mlsanity.com/paper/2305.20054/) (2023)
- [Disentangling the Impacts of Language and Channel Variability on Speech Separation Networks](https://mlsanity.com/paper/2203.16040/) (2022)
- [Reverberation as Supervision for Speech Separation](https://mlsanity.com/paper/2211.08303/) (2022)
- [On monoaural speech enhancement for automatic recognition of real noisy speech using mixture invariant training](https://mlsanity.com/paper/2205.01751/) (2022)
- [Speech Separation with Pretrained Frontend to Minimize Domain Mismatch](https://mlsanity.com/paper/2411.03085/) (2024)

## Abstract (arXiv)

The recently-proposed mixture invariant training (MixIT) is an unsupervised method for training single-channel sound separation models in the sense that it does not require ground-truth isolated reference sources. In this paper, we investigate using MixIT to adapt a separation model on real far-field overlapping reverberant and noisy speech data from the AMI Corpus. The models are tested on real AMI recordings containing overlapping speech, and are evaluated subjectively by human listeners. To objectively evaluate our models, we also devise a synthetic AMI test set. For human evaluations on real recordings, we also propose a modification of the standard MUSHRA protocol to handle imperfect reference signals, which we call MUSHIRA. Holding network architectures constant, we find that a fine-tuned semi-supervised model yields the largest SI-SNR improvement, PESQ scores, and human listening ratings across synthetic and real datasets, outperforming unadapted generalist models trained on orders of magnitude more data. Our results show that unsupervised learning through MixIT enables model adaptation on real-world unlabeled spontaneous speech recordings.