The paper investigates why automatic speech recognition (ASR) systems sometimes generate fluent text that does not correspond to the input audio, a phenomenon termed hallucination. By examining two independently trained Conformer‑Large models—one using CTC and the other RNN‑T—under conditions of environmental noise and speaker‑background shift, the authors identify the final encoder stage as a critical boundary. Bypassing this final block leads to divergence on almost all utterances, while bypassing earlier blocks has minimal effect; at this stage, representations become compact, the decoder can read the text, and grapheme information becomes explicit, yet the output is garbled or repetitive rather than fluent fabrication. The study thus pinpoints a mechanistic precondition for hallucination—failure to produce adequately grounded output—though it does not fully explain naturally occurring hallucinations, and it highlights a consistent terminal‑stage dependency across decoder families and distribution shifts.
By Hamees Sayed, Apoorv Singh, Kumar Aman, Akshat Mandloi
The paper introduces a training‑free, inference‑time technique to curb hallucinated transcripts in Whisper by projecting decoder activations away from a low‑rank hallucination subspace derived from non‑speech data. Two variants are tested: an always‑on projection that dramatically lowers hallucination rates on non‑speech benchmarks, and a gated version that applies the projection only when non‑speech is predicted, achieving a smaller but still significant reduction. On LibriSpeech, the gated method slightly increases word error rate but keeps false‑rejection rates low, demonstrating a controllable trade‑off between hallucination suppression and recognition accuracy.
By Maryam Abbasihafshejani, Murtuza Jadliwala
The paper introduces a trilingual spoken hallucination detection benchmark covering English, Russian, and Kazakh news, with 12,013 samples that include synthetic alterations and severity levels, as well as 290 fact‑checked misinformation items. Detectors are evaluated in a reference‑free setting on text, ASR transcripts, and audio, revealing that most models underperform baseline classifiers, except Gemma‑3n on transcripts. Synthetic‑trained detectors achieve high macro‑F1 scores on real‑world misinformation, but Russian provenance analysis highlights model‑dependent signals that confound synthetic benchmarks.
By Meruyert Aristombayeva, Jason S. Lucas, Chaewan Chun, Dongwon Lee
arXiv:2608. 00722v1 Announce Type: cross Abstract: Language model-based text-to-speech (LM-based TTS) remains vulnerable to speech hallucinations that deviate from the target text.
By Chenlin Liu, Minghui Fang, Zhonghao Bi, Zekai Su, Rong Wang, Jiqing Han
arXiv:2608.28916v1 Announce Type: new
Abstract: Automatic speech recognition (ASR) systems are commonly evaluated with word error rate (WER), yet many voice workflows depend on exact written values f...
By Tyler Baumgartner, Brandon Tai, Lisa Kaelin-Martin, Candice Fan, Luc Debaupte, Bill Wang, Yi Zhong
arXiv:2509.10452v3 Announce Type: replace-cross
Abstract: Pretrained automatic speech recognition (ASR) models such as Whisper perform well but still need domain adaptation to handle unseen parlance....
By Akshat Pandey, Karun Kumar, Raphael Tang
arXiv:2606. 00819v1 Announce Type: new Abstract: Large Language Models (LLMs) have achieved strong performance across diverse natural language tasks, yet their outputs often suffer from hallucinations -- content that is misaligned with factual information.
By Hanze Li, Jinhao You, Yichen Guo, Kai Tang, Shuangyang Xie, Xiande Huang
arXiv:2605.12225v3 Announce Type: replace
Abstract: While deep transformer-based models have advanced rapidly, their internal mechanisms remain largely a mystery. Recent work has prioritized understa...
By Dan Pluth, Zachary Nicholas Houghton, Yu Zhou, Vijay K. Gurbani
SPAR-K is a scheduled periodic alternating early‑exit framework for interleaved spoken language models that reduces decoding depth for speech tokens while maintaining quality. It lets most speech positions exit at a fixed intermediate layer and inserts periodic full‑depth refresh steps to counter distribution shift. Experiments on Step‑Audio‑2‑mini and GLM‑4‑Voice show up to 11 % depth reduction with less than 0.82 % drop in question‑answering accuracy and negligible impact on MOS and WER.
By Hsiao-Ying Huang, Cheng-Han Chiang, Hung-yi Lee
arXiv:2606. 07473v1 Announce Type: cross Abstract: Whisper, a widely adopted ASR model, is known to suffer from hallucinations - coherent transcriptions generated for non-speech audio entirely disconnected from the input.
By Georgii Aparin, Vadim Popov, Tasnima Sadekova, Assel Yermekova
The study investigates how textual neural models degrade when inputs contain noise such as typos, OCR errors, or dropped words. It finds that model performance decline is largely consistent across architectures under word‑level noise but diverges under character‑level noise, a difference attributed to tokenization rather than architecture. By applying a short contrastive training recipe, diverse encoders converge to a common robustness curve, enabling prediction of a model’s noise resilience and the ability to enhance robustness at specific noise scales through targeted training.
By Yefan Tao, Gerald Friedland, Luyang Kong
arXiv:2508. 07048v2 Announce Type: replace-cross Abstract: Autoregressive (AR) encoder-decoder models dominate high-quality multilingual ASR, but their left-to-right decoders make inference latency scale with transcript length.
By Taeyoun Kwon, Junhyuk Ahn, Taegeun Yun, Heeju Jwa, Yoonchae Choi, Siwon Park, Jongchan Kim, Hyungon Ryu, Hyuk-Jae Lee, Nam-Joon Kim