Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.22910

Bridge, Distill, Refine: High-Fidelity One-Step Generation for Multi-Bandwidth Speech Extension

Zhiqi Ai·Xu Shen·Mou Wang·Yongjin Zhou·Shugong Xu
View original source ↗Read source document ↗Export metadata ↓

Abstract

Speech bandwidth extension (BWE) aims to recover missing spectral content from band-limited speech. However, most generative BWE methods are designed for fixed input bandwidths and rely on iterative sampling, limiting their flexibility and efficiency. We propose BDR-BWE, a three-stage generative framework for multi-bandwidth speech extension. First, to enable conditional generation from the observed band-limited spectrum, we design a cutoff-aware conditional flow bridge in the compressed complex STFT domain, focusing generation on the missing high-frequency band while preserving observed spectral information. Second, progressive consistency distillation compresses the multi-step flow transport into a one-step generator. Third, adversarial refinement further improves high-frequency detail and harmonic coherence. BDR-BWE achieves state-of-the-art log-spectral distance (LSD) on VCTK across all evaluated bandwidths with one-step inference, while maintaining competitive perceptual quality and generalizing well to EARS. Audio samples are available at https://aizhiqi-work.github.io/bdr-bwe-demo.

Explore related subjects

eess.AS

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zhiqi Ai, Xu Shen, Mou Wang, Yongjin Zhou, Shugong Xu. 2026-10-16. Bridge, Distill, Refine: High-Fidelity One-Step Generation for Multi-Bandwidth Speech Extension. https://arxiv.org/abs/2610.22910

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

KEEP EXPLORING

Related papers

arXiv2026

Towards Prosodically Informed Mizo TTS without Explicit Tone Markings

This paper reports on the development of a text-to-speech (TTS) system for Mizo, a low-resource tonal language of the Tibeto-Burman family, spoken primarily in the Indian state of Mizoram. The TTS was built using only 5.18 hours of data; however, based on subjective and objective evaluations, the outputs were considered perceptually acceptable and intelligible. A baseline Tacotron2 model was built, and then, using the same data, another TTS model was built with VITS. VITS achieved significantly lower Mel-Cepstral Distortion (MCD) and significantly higher subjective ratings than Tacotron2, although the differences in DNSMOS, F0 RMSE, and F0 correlation were not statistically significant. In terms of tone synthesis, the VITS model showed numerically lower tone errors than the Tacotron2 model. The paper demonstrates that a non-autoregressive, end-to-end framework can achieve synthesis of acceptable perceptual quality and intelligibility.

eess.AS↗
arXiv2026

AcousticAgent: A Floorplan-Centric Simulation Platform for Distributed Multi-Room Speech Processing

Existing acoustic simulation tools either generate room impulse responses (RIRs) efficiently in parameterized rooms or rely on preconstructed three-dimensional scenes, making it difficult to build large numbers of editable residential multi-room environments. We present AcousticAgent, a floorplan-centric acoustic simulation platform that converts structured floorplans into editable, acoustically connected multi-room scenes. AcousticAgent jointly models room geometry and connectivity, furniture, materials, sound sources, and microphone configurations. A scene-oriented Acoustic Materials Database maps semantic elements to acoustic parameters, while a floorplan-aware hybrid geometrical-acoustics solver generates multichannel RIRs and reverberant audio. The platform also supports directional and moving sources, distributed microphones, and VLM-assisted scene construction from floorplan images and text. Using AcousticAgent, we construct FP-RIR, a dataset covering 15,376 residential floorplans with diverse room topologies, propagation conditions, and source--receiver configurations. We validate its acoustic fidelity against real-world measurements and evaluate whole-home source localization, speech enhancement, and far-field speech processing under single-device and distributed microphone settings. The results demonstrate the effectiveness of floorplan-aware modeling for large-scale residential acoustic simulation and the benefits of distributed spatial processing in multi-room environments. AcousticAgent and FP-RIR provide scalable and configurable resources for developing and evaluating residential speech systems.

eess.AS↗
arXiv2026

VoxTurn: Answering Several Turn-Taking Questions in One Pass with One Model

A cascaded voice agent must decide, throughout each silence of its user, whether the user has finished their turn: answering too early cuts the user off, answering too late makes the agent slow. It often needs related decisions too: will the user soon go on, is speech during its own turn an interruption or a backchannel, is a sound speech at all. With open detectors, each extra decision costs another model or another call, and on the 14-language eot-bench they cut users off in more than a quarter of mid-turn pauses. Because they are trained on segmented utterances, this mismatch is the largest single source of error. We present VoxTurn, a 0.8B speech LLM that casts these decisions as multiple-choice questions, scores the hidden state of each option, and answers any set of them in one forward pass with a block-isolating attention mask, exactly as if asked one at a time. Trained on real conversational pauses with labels derived from their timing, and re-queried while a silence lasts, VoxTurn reaches 11.5% false cut-offs in English at a 300 ms latency budget, comparable to two commercial detectors, and a 14-language mean of 19.9%, the lowest among open models, at 8.3 ms per decision. In Pipecat and LiveKit it cuts off 17% and 9% of turns, and it ranks fifth of 24 in TurnBench end-of-turn detection without any TurnBench training data. The model, code and framework plugins are released.

eess.AS↗