Daily Papers — 2026-07-26

7 papers on audio, speech, music, and acoustics.

1. Indic DiarBench: A Multilingual Joint Diarization and ASR Benchmark for Indian Languages

Authors: Deovrat Mehendale, Aditya Mehndiratta, Dhruv Rathi, Kaushal Bhogale, Mitesh M. Khapra

Categories: cs.CL, cs.AI | 5 pages, 2 figures, Interspeech 2026 conference Score: 7.42/10 (Obj:9 Id:9 Ind:6 Comp:8 Eff:8 Nov:8 Rep:5)

Strength: 首个覆盖全部 22 种印度scheduled语言的联合日志化-ASR 基准（108h, 485 说话人, 3 种声学条件），以 CC BY 4.0 开放发布，填补了多语种多语者语音处理评测的关键空白。
Weakness: 基线评测依赖商业 API（版本不锁定导致不可复现），最优系统 Sarvam 为作者机构产品且内部不透明，标注一致性未量化，方法层面无新颖性。
Full review: Claude Code 全文七公理审稿

In this work, we introduce Indic DiarBench, a speaker diarization and ASR benchmark dataset spanning all 22 scheduled languages of India. This corpus comprises approximately 108 hours of natural multi-speaker audio from near-field meetings, far-field recordings, and in-the-wild audios. All annotations are human-corrected with time-aligned speaker attributed transcriptions. The dataset captures conversational nuance prevalent in Indian speech, such as English code-mixing, dialectal variation, and frequent speaker overlap. To establish a baseline for joint ASR and diarization capabilities we evaluate leading systems including commercial speech APIs and multimodal large language models. Indic DiarBench is released as an open-access resource to advance inclusive, multilingual speech technology research for Indian languages.

Authors: Jun Zhan, Chen Yang, Yitian Gong, Donghua Yu, Kuangwei Chen et al.

Categories: cs.SD, cs.CV | 15 pages, 2 figures, 6 tables Score: 6.42/10 (Obj:8 Id:8 Ind:5 Comp:8 Eff:6 Nov:5 Rep:9)

Strength: 通过 progressive ablation 干净地证明 segment-level contrastive alignment 与 modality-specific semantic distillation 互补——前者把 Desync 从 0.884 降到 0.570，后者把 WER 从 0.243 降到 0.205，组合在 Verse-Bench T2AV 上全面最优，且两目标训练-only、零推理开销。
Weakness: 下游效用只在四个内部 VAE 配置间比较，未与同期强系统（UniAVGen CVPR2026 / AVTok ECCV2026 / AV-Link ICCV2025）在统一 benchmark 上 head-to-head；核心新颖性是 Synchformer contrastive + REPA distillation 的跨模态组合，且 concurrent AVTok 削弱 “first audio-video VAE for c。
Full review: Claude Code 全文七公理审稿

Recent generative models are moving beyond silent video or standalone audio synthesis toward the joint generation of synchronized audio and video. Despite this progress, jointly generating audio and video with fine-grained cross-modal correspondence remains challenging due to their fundamental structural differences. Most existing methods use audio and video VAEs trained separately. As a result, the two latent spaces lack cross-modal alignment, leaving the downstream generative model to learn cross-modal synchronization from scratch. We present OmniVAE, a jointly trained audio-video VAE that learns fine-grained semantic alignment between audio and video latent representations. Beyond reconstruction, OmniVAE uses a segment-level audio-video contrastive objective to capture temporal-semantic correspondence and align the two latent spaces. In parallel, it distills features from pretrained modality-specific semantic encoders into each modality, improving the downstream learnability of both latent spaces. Extensive experiments show that both objectives consistently improve the learnability of the latent spaces, translating into higher generation quality and more accurate cross-modal synchronization in downstream text-to-audio-video generation. These findings underscore the importance of learning unified representations as a foundation for omnimodal modeling.1

3. Expose Your Disguise: Recovering Source Speaker Identity From Voice Conversion

Authors: Hanlei Zhang, Zhongming Ma, Mingyang Zhang, Tengfei Liu, Yushi Cheng et al.

Categories: cs.SD, eess.AS Score: 6.32/10 (Obj:8 Id:6 Ind:5 Comp:5 Eff:8 Nov:5 Rep:8)

Strength: TRIDENT 在 7 种主流 VC 方法上实现 90.99% Top-1 源说话人恢复准确率，较 SOTA Revelio 提升 12.08pp，且在 telephony/VoIP/cross-lingual/adaptive 对手场景下保持鲁棒性，代码和数据承诺开源。
Weakness: 方法本质是 ECAPA-TDNN + VC 分类器 + 目标 speaker 分支的工程组合，缺少最简 baseline 和参数量对齐的消融，unseen VC 绝对准确率偏低（AdaIN Top-1 仅 61.57%），训练-评测数据同源且 disentanglement 标签来自自建生成管线，独立性不足。
Full review: Claude Code 全文七公理审稿

Voice conversion (VC) poses a significant threat to biometric security by allowing attackers to impersonate target speakers. In forensic contexts, recovering the source speaker’s identity from converted audio is vital for narrowing the field of suspects. To address this, we propose TRIDENT, a retracing framework designed to restore a source speaker’s original identity from a converted audio sample. TRIDENT utilizes a three-pronged architecture consisting of a primary extractor and two auxiliary branches. The first auxiliary branch identifies the underlying voice conversion mechanism. This design acknowledges that even if the exact conversion strategy is unknown, a high-performance model adopted by the attacker is typically a derivative or variant of established mainstream ones. The second auxiliary branch extracts a latent representation of the target speaker, facilitating the isolation of target-specific traits from the composite converted audio sample. Finally, the main extractor leverages insights from both auxiliary branches to decouple confounding factors and distill a highly discriminative representation of the source speaker’s identity. Experimental results demonstrate that TRIDENT achieves an accuracy as high as 90.99% against 7 state-of-the-art voice conversion methods. Furthermore, TRIDENT maintains robust performance under challenging conditions, including telephony channels, unseen languages, and adaptive scenarios.

4. Automatic Audio Equalization with Semantic Embeddings

Authors: Eloi Moliner, Vesa Välimäki, Konstantinos Drossos, Matti S. Hämäläinen

Categories: cs.SD, eess.AS | Presented at AES International Conference on Artificial Intelligence and Machine Learning for Audio. London, UK. 2025 Score: 5.79/10 (Obj:8 Id:5 Ind:8 Comp:8 Eff:5 Nov:5 Rep:6)

Strength: 方法极简（0.34M 可训练参数 + frozen CLAP），音乐 EQ 场景主观测试与 Oracle 无显著差异（p=0.72），证实轻量 head 足以估计目标 log-mel 谱。
Weakness: 语音 EQ 场景相对最简 Average baseline 无显著提升，带噪/带混响场景被前端 DeepFilterNet2 掩盖，且未与 diffusion EQ [13]、Peladeau 2024 [10] 等近期强 baseline 直接对比。
Full review: Claude Code 全文七公理审稿

This paper presents a data-driven approach to automatic blind equalization of audio by predicting log-mel spectral features and deriving an inverse filter. The method uses a deep neural network, where a pre-trained model provides semantic embeddings as a backbone, and only a lightweight head is trained. This design is intended to enhance training efficiency and generalization. Trained on both music and speech, the model is robust to noise and reverberation. Objective evaluations confirm its effectiveness, and subjective tests show performance comparable to that of an oracle that uses true log-mel spectral features, indicating that the model accurately estimates the desired characteristics, with remaining limitations attributed to the filtering stage. Overall, the results highlight the potential of the method for real-world audio enhancement applications.

5. Improving Zero-Shot Phonetic Classification through Language-Agnostic Articulatory Features

Authors: Ryo Magoshi, Jaeyoung Lee, Shinsuke Sakai, Tatsuya Kawahara

Categories: cs.SD, eess.AS | Accepted at Interspeech 2026 Score: 5.79/10 (Obj:8 Id:5 Ind:8 Comp:6 Eff:5 Nov:4 Rep:6)

Strength: AF-based 分类在 Japanese nasal [ɴ] recall 上从 3.1% 提升至 38.5%（segmental AF），并揭示 single-frame vs segmental 聚合应匹配 cue 的时域特性（aspiration 用单帧，nasal 用段级）。
Weakness: 遗漏 Allophant (2023) 和 Allosaurus 两个使用 articulatory features 的核心 zero-shot baseline，且未隔离 XLS-R 预训练与 AFCM 的贡献，nasal 分类绝对准确率仅 72.6% 仍未达可用水平。
Full review: Claude Code 全文七公理审稿

Recent Phonetic Foundation Models (PFMs) for Speech-to-IPA transcription rely on Grapheme-to-Phoneme (G2P) labels, but the phoneme labels are not necessarily phonetically faithful. To investigate this issue, we evaluate zero-shot phonetic classification on Chinese aspiration and Japanese moraic nasals. A PFM trained on G2P-labeled data excluding these two languages yields poor accuracy on both tasks, showing that multilingual coverage with discrete IPA tokens is not sufficient for unseen settings. To overcome this limitation, we propose a classification method based on continuous Articulatory Feature (AF) vectors extracted from each frame. This AF-based approach outperforms discrete token-based methods, particularly for rare phones. We further show that it is crucial to adopt the optimal temporal aggregation of AF vectors for the target distinction: single-frame classification is best for aspiration, while segmental classification substantially improves nasal classification.

6. Earnings25: A Comprehensive 500-Hour Speech Benchmark for Finance

Authors: Denglin Jiang, Haoran Zhou, Anshul Wadhawan, Brendan Fahy, Vinay Ramesh et al.

Categories: cs.CL, cs.AI | 5 pages, 0 figures, 5 tables Score: 5.47/10 (Obj:8 Id:5 Ind:8 Comp:5 Eff:5 Nov:4 Rep:8)

Strength: 498h 真实收益电话 + industry-stratified 46h 分层评测集，配 speaker/industry 元数据与可复现基线（Parakeet-TDT 最低 WER 10.8%），填补了金融 ASR benchmark 的 metadata 缺口。
Weakness: Baseline 覆盖仅 4 个模型配置，遗漏 Whisper-v3/Canary/SPGISpeech-finetuned/commercial API，且未引用早 9 个月的 Open ASR Leaderboard (2510.06961) 和早 11 个月的 SPGISpeech 2.0 (2508.05554)；声称 speaker-aware 评估却未报告 speaker-role 分层 WER；Earnings-22 时长。
Full review: Claude Code 全文七公理审稿

We introduce Earnings25, a finance-domain benchmark for evaluating automatic speech recognition (ASR) on English-language earnings calls under realistic conditions. Earnings25 comprises two complementary test sets: (i) testset-full, 498 hours of full English-language S&P 500 earnings calls from Q4 2025, and (ii) testset-segmented, a 46-hour industry-balanced set of 290 segments sampled from English-language U.S. earnings calls in 2025. The benchmark provides aligned transcripts and structured metadata, including speaker roles, industry labels, and call structure, enabling speaker- and industry-aware evaluation beyond aggregate word error rate (WER). We report reproducible baselines for Whisper and Parakeet-TDT using standardized scoring.

7. Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers

Authors: Dongseong Hwang, Prasanth Yadla, Kaan Elgin, Shifas Padinjaru Veettil, Sivanand Achanta et al.

Categories: cs.SD, cs.CL | 11 pages, ICASSP Score: 5.40/10 (Obj:9 Id:5 Ind:5 Comp:8 Eff:6 Nov:5 Rep:2)

Strength: 架构实现真正的压缩——单个全共享深度解码器通过 DiT 阶段条件（RoPE 编码码本索引）取代 Moshi 的每码本投影，配合固定窗口 KV 缓存实现常数内存（21 MB 峰值，20-320s 不变），在 AMX 上 16× 实时运行，生产 MOS 4.15（+0.28 over baseline）。
Weakness: 核心消融虽隔离变量但基线比较不公平（Moshi 运行在 12.5Hz/1.1kbps vs 本文 25Hz/1.5kbps，未提供 Moshi 的 scaling 对比），生产 MOS +0.28 无法归因到 detokenizer（FM 本身也变了），且训练数据/代码/模型/AMX 硬件全部闭源，结果不可独立复现。
Full review: Claude Code 全文七公理审稿

Siri Expressive Voices synthesize rich, configurable speech in real time and entirely on device, powered by AFM 3 Core Advanced, Apple’s most powerful on-device foundation model. This work presents the memory-efficient audio synthesis architecture behind that capability: a detokenizer that converts the semantic audio tokens emitted by the foundation model into high-fidelity audio within the tight compute and memory budget of the Apple Matrix Coprocessor (AMX). We convert semantic audio tokens to a residual vector quantization (RVQ) representation with a three-component design, a streaming encoder, a temporal decoder, and a depth decoder, that systematically decouples temporal and depth processing. A single reusable depth decoder with Diffusion Transformer (DiT)-style stage conditioning generates all RVQ levels autoregressively, replacing the dedicated per-level decoders of prior multi-decoder architectures, while causal sliding window attention with fixed-window key-value caching yields constant memory complexity independent of sequence length. Deployed on the AMX, the detokenizer sustains roughly 10 ms per generation step, about 16x faster than real time, with a peak runtime memory of only 21 MB and 329 MB of on-device assets, enabling continuous streaming synthesis of 20-320 seconds of audio. This constant, small footprint replaces the linear and quadratic memory scaling of conventional transformer- and GAN-based approaches. Ablation studies validate the key architectural components, and audio quality assessment confirms that the architecture maintains synthesis fidelity while achieving efficiency gains over existing methods. Operating at a 1-billion-parameter activation size within AFM 3 Core Advanced, it improves Mean Opinion Score by +0.28 overall (4.15 vs. 3.87) and by +0.42 on conversational speech (4.24 vs. 3.82) over the prior on-device text-to-speech system.