Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios

Changhao Pan; Rui Yang; Han Wang; Zhuan Zhou; Xuming He; Wenxiang Guo; Ziyue Jiang; Ruiqi Li; Yu Zhang; Chenyuhao Wen; Ke Lei; Xiang Yin; Jingyu Lu; Zhiyuan Zhu; Zhou Zhao

arXiv · 2605.28618

Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios

Changhao Pan·Rui Yang·Han Wang·Zhuan Zhou·Xuming He·Wenxiang Guo·Ziyue Jiang·Ruiqi Li·Yu Zhang·Chenyuhao Wen·Ke Lei·Xiang Yin·Jingyu Lu·Zhiyuan Zhu·Zhou Zhao

View original source ↗Read source document ↗Export metadata ↓

Abstract

Recent advances in speech generation have enabled high-fidelity synthesis, yet systematic evaluation of models under long-context conditions remains largely underexplored. A comprehensive evaluation benchmark for long-form speech is indispensable for two reasons: 1) existing test scenarios are often confined to limited domains, creating a significant gap with the diverse downstream applications; 2) existing metrics overlook critical long-text factors such as consistency and coherence, failing to generalize reliably. To this end, we propose Swanbench-Speech, a comprehensive benchmark that decomposes long-form speech quality into specific, disentangled dimensions. SwanBench-Speech has three key properties. 1) Rich speech scenarios: Focusing on long-form speech generation and dialog generation, SwanBench-Speech covers acoustics, semantics, and expressiveness challenges, and consists of 1,101 samples spanning 17 common speech scenarios; 2) Comprehensive evaluation dimensions: Along the acoustics, semantics, and expressiveness axes, SwanBench-Speech defines an automated evaluation protocol with seven metrics to provide a comprehensive, accurate, and standardized assessment; 3) Valuable Insights: Through extensive experiments, we reveal that current models still struggle in highly expressive scenarios and exhibit a notable gap in consistency and hierarchy compared to real recordings.

Explore related subjects

eess.AS

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Changhao Pan, Rui Yang, Han Wang, Zhuan Zhou, Xuming He, Wenxiang Guo, Ziyue Jiang, Ruiqi Li, Yu Zhang, Chenyuhao Wen, Ke Lei, Xiang Yin, Jingyu Lu, Zhiyuan Zhu, Zhou Zhao. 2026-05-27. Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios. https://arxiv.org/abs/2605.28618

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

KEEP EXPLORING

Related papers

arXiv2026

LLMs and Speech: Integration vs. Combination

In this work, we study different approaches to utilize large language models (LLMs) for automatic speech recognition (ASR). Specifically, we compare the tight integration of an acoustic model (AM) with the LLM ("speech LLM") to the traditional way of combining AM and LLM via shallow fusion and provide ablations on the effect of different label units and LLM sizes. For tight integration, we further examine the effect of attention interfaces, encoder downsampling, and length normalization. Furthermore, we investigate joint recognition with a CTC model to mitigate hallucinations of speech LLMs and present effective optimizations. We train and evaluate on LibriSpeech and Loquacious and additionally evaluate on the HuggingFace ASR leaderboard. Across model sizes, we find that shallow fusion consistently outperforms tight integration of AM and LLM on in-domain data, highlighting the importance of strong shallow-fusion baselines when evaluating speech LLMs for ASR. On the more heterogeneous HuggingFace ASR leaderboard, however, the integrated prefix LLM achieves lower average WER than shallow fusion, with gains concentrated on out-of-domain corpora.

eess.AS↗

arXiv2026

Vaani Benchmark V1.0: An Inclusive Multimodal Benchmark Dataset for Hindi

Benchmarking is critical for the systematic evaluation of machine learning systems. While several open-source datasets are available for Hindi, existing benchmarks remain limited in terms of modality, geographic diversity, demographic representation, and transcription robustness. We introduce an inclusive, multimodal Hindi benchmark dataset collected from 102 districts across India. The dataset consists of spontaneous speech elicited using image prompts and recorded under real-world acoustic conditions across diverse demographic groups. Each audio segment is associated with the image prompt that elicited it and is annotated with three independent transcriptions, enabling multi-reference evaluation that accounts for permissible orthographic and lexical variations. We show that single-reference evaluation overstates errors, while multi-reference evaluation provides a more robust, inclusive, and realistic assessment of automatic speech recognition (ASR) systems. The pairing of each utterance with its eliciting image further enables image retrieval evaluation using both speech and text queries. We evaluate multiple multimodal embedding models for image retrieval and report their performance on the combined speech--image and text--image retrieval tasks. The results reveal a performance gap between text-based and audio-based retrieval, with text-based retrieval consistently achieving higher performance.

eess.AS↗

arXiv2026

Cover First, Disagree Softly: Rethinking Mismatch-First Active Learning for Frame-Level Audio Classification

Sound event detection relies on frame-level strong labels whose annotation is expensive. Active learning addresses this problem by selecting the audio segments whose labels help the classifier most. One of the prevailing acquisition strategies for this task, mismatch-first farthest-traversal (MFFT), combines the disagreement between two classifiers and the diversity of the selected segments through hard sequential decisions. It selects whole groups of high-disagreement segments first and spreads only the remaining budget by farthest traversal. On two multi-label datasets we show that this design is blind to the similarity among the selected segments and fails under low budgets, with every mismatch-first variant ending below the plain geometric strategy it builds on. We propose mismatch-weighted facility location (MW-FL), which spends the entire budget through a disagreement-weighted coverage objective that penalizes similarity among the selected segments. The disagreement signal from MFFT is used to obtain the nonnegative weights of this facility-location objective, using fixed smoothing without dataset-specific tuning. Experiments across two geometric mechanisms with three ways of using disagreement show that coverage of the selected segments is the dominant factor, hard disagreement gating of selection is harmful on both mechanisms, and soft disagreement weighting helps on top of coverage. MW-FL attains the best area under the learning curve on both datasets.

eess.AS↗