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

Experiment configuration. We train a separation model f using either one or both datasets paired with appropriate loss functions: supervised PIT with reverberant Libri2Mix (RL2M) or unsupervised MixIT with AMI. Model parameters θ can either be randomly initialized or warm-started from another model pretrained with unsupervised MixIT on AudioSet.

Abstract

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.

Paper

Audio Demos

Dataset Recipe

Last updated: May 2022