145 papers
This paper introduces Geo-ATBench, a new benchmark and the Geo-AT task that leverage geospatial semantic context to resolve acoustic ambiguities in multi-label audio tagging, demonstrating through the GeoFusion-AT framework that incorporating location-based priors significantly improves recognition performance and aligns with human judgment.
The paper proposes G-STAR, an end-to-end system that integrates a time-aware speaker-tracking module with a Speech-LLM backbone to achieve robust, timestamped speaker-attributed recognition for long-form, overlapping multi-party speech while maintaining global identity consistency.
The paper introduces FireRedASR2S, a state-of-the-art industrial-grade all-in-one automatic speech recognition system that unifies high-performance modules for speech transcription, voice activity detection, language identification, and punctuation prediction, achieving superior results across Mandarin, Chinese dialects, and English benchmarks compared to existing solutions.
The paper proposes CSP-FT, a characteristic-specific partial fine-tuning strategy that selectively updates only the most and least relevant layers of LLM-based TTS models to achieve superior emotion and speaker adaptation with significantly faster training and reduced catastrophic forgetting compared to full fine-tuning.
This paper introduces HVAC-EAR, a novel system that reconstructs intelligible human speech from low-resolution, noisy pressure data in HVAC systems using a complex-valued conformer, demonstrating significant privacy risks by achieving clear eavesdropping up to 1.2 meters in real-world deployments.
This paper introduces ML-ITW, a large-scale multilingual dataset designed to evaluate speech deepfake detectors in real-world conditions, revealing that current detection methods suffer significant performance degradation when facing diverse languages and platform-specific compression artifacts.
This paper proposes an embedding-based data selection strategy that leverages speaker, phonetic, and semantic features to identify a small, high-quality subset of in-the-wild speech data, demonstrating that training on just 5% of the original 100k-hour dataset can significantly outperform full-dataset training for domain-specific speech recognition tasks.
This paper presents an empirical study on adapting speech foundation models to low-resource Pacific Indigenous languages, revealing that while strategies like Low-Rank Adaptation offer initial success, they ultimately struggle with catastrophic forgetting and internal representational drift during sequential learning, highlighting the urgent need for robust adaptation frameworks that balance plasticity and stability.
This paper proposes a compact acoustic framework that combines multi-branch CNN feature extraction with an efficient Legendre Memory Unit (LMU) for temporal modeling and a calibrated posterior ensemble fusion strategy to achieve robust, real-time cross-domain infant cry classification despite limited annotations and strong domain shifts.
This paper introduces the Deep Koopman Speech Disentanglement Autoencoder (DKSD-AE), a scalable and efficient architecture that leverages Koopman operators and instance normalization to effectively disentangle speaker identity from linguistic content for robust speaker verification without relying on textual supervision or large pretrained models.
This paper introduces Whisper-RIR-Mega, a benchmark dataset pairing clean LibriSpeech utterances with real room impulse responses to evaluate and demonstrate the performance degradation of various Whisper models under reverberant conditions, while providing open-source resources for reproducible research on robust ASR.
This paper proposes a novel end-to-end audio-visual speech recognition framework that integrates speech enhancement via a Conformer-based bottleneck fusion module to implicitly refine noisy audio features without explicit mask generation, thereby preserving semantic integrity and outperforming existing mask-based methods on the LRS3 benchmark under noisy conditions.
RAMoEA-QA is a hierarchically routed generative model that employs a two-stage conditional specialization mechanism—combining an Audio Mixture-of-Experts for acoustic encoding and a Language Mixture-of-Adapters for query intent—to achieve state-of-the-art robustness and accuracy in respiratory audio question answering across diverse devices, environments, and task shifts.
This paper proposes a prosodic-boundary-aware post-training strategy for LLM-based TTS that enables natural streaming generation with incremental text input by learning early stopping at content boundaries and utilizing a sliding-window prompt to prevent long-form collapse, significantly outperforming existing baselines in both short and long-form scenarios.
Whisper-CD is a training-free, inference-time contrastive decoding framework that mitigates hallucinations and repetition in long-form speech recognition by contrasting clean audio logits against a unified objective derived from multiple acoustically motivated negative perturbations, thereby significantly reducing word error rates and improving generation throughput without requiring model retraining.
This paper introduces RAPTOR, a controlled study demonstrating that multilingual HuBERT pre-training, rather than model scale, is the primary driver of cross-domain robustness and reliable calibration in compact audio deepfake detection systems.
TempoSyncDiff is a reference-conditioned latent diffusion framework that employs teacher-student distillation and temporal regularization to enable low-latency, temporally stable, and identity-consistent audio-driven talking head generation suitable for edge deployment.
The paper introduces Omni-C, a single dense Transformer encoder that compresses heterogeneous modalities (text, audio, and image) into shared representations via unimodal contrastive pretraining, thereby eliminating the parameter overhead and routing complexity of Mixture-of-Expert architectures while achieving comparable performance with significantly reduced memory usage.
This paper demonstrates that a lightweight configuration of the pre-trained XLS-R model, utilizing only its first three layers, achieves state-of-the-art cough activity detection for automatic tuberculosis screening, significantly outperforming existing baselines while offering the computational efficiency required for smartphone-based deployment.
This paper investigates the use of a text-trained large language model adapted for speech to localize fake words in partially edited audio, revealing that while the model effectively identifies edits by leveraging specific training patterns like word-level polarity substitutions, it struggles to generalize to unseen editing styles.