SPEAR-Gen: Generation-Aware Pre-training for Unified Speech Representations

Xiaoyu Yang; Arthur Hinsvark; Antonios Alexos; Osama Hanna; Philip C. Woodland; Yiting Lu

Home/Archive/arXiv

arXiv · 2609.34147

SPEAR-Gen: Generation-Aware Pre-training for Unified Speech Representations

Xiaoyu Yang·Arthur Hinsvark·Antonios Alexos·Osama Hanna·Philip C. Woodland·Yiting Lu

View original source ↗Read source document ↗Export metadata ↓

Abstract

Speech understanding and generation place different demands on speech representations, and existing models are typically optimised towards one capability or the other. To reduce this gap, we introduce SPEAR-Gen, a speech representation model that learns a single representation for both capabilities. Task-aligned feature aggregation consolidates complementary linguistic and paralinguistic information across a frozen encoder into discrete targets for masked prediction, while a coarse-to-fine objective combines log-Mel reconstruction with residual flow matching to preserve spectral structure and fine-grained acoustic variation. Experiments on SUPERB and speech resynthesis show that SPEAR-Gen maintains strong understanding performance while substantially improving resynthesis quality and speaker preservation. These results demonstrate that a single speech representation can effectively support both understanding and generation.

Explore related subjects

eess.AS cs.SD

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Xiaoyu Yang, Arthur Hinsvark, Antonios Alexos, Osama Hanna, Philip C. Woodland, Yiting Lu. 2026-09-28. SPEAR-Gen: Generation-Aware Pre-training for Unified Speech Representations. https://arxiv.org/abs/2609.34147

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

TACO: Training-free Sound Prompted Segmentation via Semantically Constrained Audio-visual CO-factorization

Large-scale pre-trained audio and image models demonstrate an unprecedented degree of generalization, making them suitable for a wide range of applications. Here, we tackle the specific task of sound-prompted segmentation, aiming to segment image regions corresponding to objects heard in an audio signal. Most existing approaches tackle this problem by fine-tuning pre-trained models or by training additional modules specifically for the task. We adopt a different strategy: we introduce a training-free approach that leverages Non-negative Matrix Factorization (NMF) to co-factorize audio and visual features from pre-trained models so as to reveal shared interpretable concepts. These concepts are passed on to an open-vocabulary segmentation model for precise segmentation maps. By using frozen pre-trained models, our method achieves high generalization and establishes state-of-the-art performance in unsupervised sound-prompted segmentation, significantly surpassing previous unsupervised methods.

eess.AS↗

arXiv2026

ImmersiveFlow: Stereo-to-7.1.4 spatial audio generation with flow matching

Immersive spatial audio has become increasingly critical for applications ranging from AR/VR to home entertainment and automotive sound systems. However, existing generative methods remain constrained to low-dimensional formats such as binaural audio and First-Order Ambisonics (FOA). Binaural rendering is inherently limited to headphone playback, while FOA suffers from spatial aliasing and insufficient resolution for high-frequency. To overcome these limitations, we introduce ImmersiveFlow, the first end-to-end generative framework that directly synthesizes discrete 7.1.4 format spatial audio from stereo input. ImmersiveFlow leverages Flow Matching to learn trajectories from stereo inputs to multichannel spatial features within a pretrained VAE latent space. At inference, the Flow Matching model predicted latent features are decoded by the VAE and converted into the final 7.1.4 waveform. Comprehensive objective and subjective evaluations demonstrate that our method produces perceptually rich sound fields and enhanced spatial impression, significantly outperforming traditional upmixing techniques.

eess.AS↗

arXiv2026

Data Augmentation for L2 English Speaking Assessment using TTS

Automated assessment of second language (L2) speaking proficiency requires substantial annotated speech data, which are scarce compared to written learner corpora. We investigate whether written L2 responses can be transformed into useful synthetic speech for proficiency assessment using text-to-speech (TTS) and voice cloning. Using COREFL, a corpus of paired spoken and written responses from L2 learners of English, we systematically study two factors: how written responses should be transformed into spoken-style language ("speechification") and how synthetic voices and texts should be paired based on shared learner attributes (proficiency level, first language, both, or neither). We generate speechified responses with a large language model and synthesise them using TTS and voice cloning, then evaluate their utility for audio-based (HuBERT) and text-based (ModernBERT) proficiency grading. Results show that pairing voices and texts based on proficiency provides a principled strategy for synthetic data generation, while speechification substantially improves the match between written and spoken L2 and improves downstream grading performance. Augmenting real training data with synthetic speechified responses improves both HuBERT- and ModernBERT-based graders, demonstrating the potential of synthetic spoken data for L2 speaking proficiency assessment.

eess.AS↗