Qwen3-ASR Technical Report

In this report, we introduce Qwen3-ASR family, which includes two powerful all-in-one speech recognition models and a novel non-autoregressive speech forced alignment model. Qwen3-ASR-1.7B and Qwen3-ASR-0.6B are ASR models that support language identification and ASR for 52 languages and dialects. Both of them leverage large-scale speech training data and the strong audio understanding ability of their foundation model Qwen3-Omni. We conduct comprehensive internal evaluation besides the open-sourced benchmarks as ASR models might differ little on open-sourced benchmark scores but exhibit significant quality differences in real-world scenarios. The experiments reveal that the 1.7B version achieves SOTA performance among open-sourced ASR models and is competitive with the strongest proprietary APIs while the 0.6B version offers the best accuracy-efficiency trade-off. Qwen3-ASR-0.6B can achieve an average TTFT as low as 92ms and transcribe 2000 seconds speech in 1 second at a concurrency of 128. Qwen3-ForcedAligner-0.6B is an LLM based NAR timestamp predictor that is able to align text-speech pairs in 11 languages. Timestamp accuracy experiments show that the proposed model outperforms the three strongest force alignment models and takes more advantages in efficiency and versatility. To further accelerate the community research of ASR and audio understanding, we release these models under the Apache 2.0 license.

SLM-SS: Speech Language Model for Generative Speech Separation

Speech separation (SS) has advanced significantly with neural network-based methods, showing improved performance on signal-level metrics. However, these methods often struggle to maintain speech intelligibility in the separated signals, which can negatively affect the performance of downstream tasks such as speech recognition. In this work, we propose SLM-SS, a novel approach that applies speech language models to SS, aiming to enhance the intelligibility and coherence of the separated signals. We frame SS as discrete multi-codebook sequence generation, using Encoder-Decoder models to map quantized speech mixtures to target tokens. In addition to the autoregressive modeling strategy, we introduce a non-autoregressive model to improve decoding efficiency for residual tokens. Experimental results on the LibriMix dataset demonstrate that our approach shows significantly better preservation of speech intelligibility, leading to improved linguistic consistency in a variety of downstream tasks compared to existing approaches.

Distillation-based Layer Dropping (DLD): Effective End-to-end Framework for Dynamic Speech Networks

Edge devices operate in constrained and varying resource settings, requiring dynamic architectures that can adapt to limitations of the available resources. To meet such demands, layer dropping ($\mathcal{LD}$) approach is typically used to transform static models into dynamic ones by skipping parts of the network along with reducing overall computational complexity. However, existing $\mathcal{LD}$ methods greatly impact the dynamic model's performance for low and high dropping cases, deteriorating the performance-computation trade-off. To this end, we propose a distillation-based layer dropping (DLD) framework that effectively combines the capabilities of knowledge distillation and $\mathcal{LD}$ in an end-to-end fashion, thereby achieving state-of-the-art performance for dynamic speech networks. Comprehensive experimentation utilizing well-known speech recognition methods, including conformer and WavLM, on three public benchmarks demonstrates the effectiveness of our framework, reducing the word error rate by $9.32\%$ and $2.25\%$ for high and no dropping cases with $33.3\%$ reduction in training time.

SE-DiCoW: Self-Enrolled Diarization-Conditioned Whisper

Speaker-attributed automatic speech recognition (ASR) in multi-speaker environments remains a major challenge. While some approaches achieve strong performance when fine-tuned on specific domains, few systems generalize well across out-of-domain datasets. Our prior work, Diarization-Conditioned Whisper (DiCoW), leverages speaker diarization outputs as conditioning information and, with minimal fine-tuning, demonstrated strong multilingual and multi-domain performance. In this paper, we address a key limitation of DiCoW: ambiguity in Silence-Target-Non-target-Overlap (STNO) masks, where two or more fully overlapping speakers may have nearly identical conditioning despite differing transcriptions. We introduce SE-DiCoW (Self-Enrolled Diarization-Conditioned Whisper), which uses diarization output to locate an enrollment segment anywhere in the conversation where the target speaker is most active. This enrollment segment is used as fixed conditioning via cross-attention at each encoder layer. We further refine DiCoW with improved data segmentation, model initialization, and augmentation. Together, these advances yield substantial gains: SE-DiCoW reduces macro-averaged tcpWER by 52.4% relative to the original DiCoW on the EMMA MT-ASR benchmark.

Multilingual Dysarthric Speech Assessment Using Universal Phone Recognition and Language-Specific Phonemic Contrast Modeling

The growing prevalence of neurological disorders associated with dysarthria motivates the need for automated intelligibility assessment methods that are applicalbe across languages. However, most existing approaches are either limited to a single language or fail to capture language-specific factors shaping intelligibility. We present a multilingual phoneme-production assessment framework that integrates universal phone recognition with language-specific phoneme interpretation using contrastive phonological feature distances for phone-to-phoneme mapping and sequence alignment. The framework yields three metrics: phoneme error rate (PER), phonological feature error rate (PFER), and a newly proposed alignment-free measure, phoneme coverage (PhonCov). Analysis on English, Spanish, Italian, and Tamil show that PER benefits from the combination of mapping and alignment, PFER from alignment alone, and PhonCov from mapping. Further analyses demonstrate that the proposed framework captures clinically meaningful patterns of intelligibility degradation consistent with established observations of dysarthric speech.

MA-LipNet: Multi-Dimensional Attention Networks for Robust Lipreading

Lipreading, the technology of decoding spoken content from silent videos of lip movements, holds significant application value in fields such as public security. However, due to the subtle nature of articulatory gestures, existing lipreading methods often suffer from limited feature discriminability and poor generalization capabilities. To address these challenges, this paper delves into the purification of visual features from temporal, spatial, and channel dimensions. We propose a novel method named Multi-Attention Lipreading Network(MA-LipNet). The core of MA-LipNet lies in its sequential application of three dedicated attention modules. Firstly, a \textit{Channel Attention (CA)} module is employed to adaptively recalibrate channel-wise features, thereby mitigating interference from less informative channels. Subsequently, two spatio-temporal attention modules with distinct granularities-\textit{Joint Spatial-Temporal Attention (JSTA)} and \textit{Separate Spatial-Temporal Attention (SSTA)}-are leveraged to suppress the influence of irrelevant pixels and video frames. The JSTA module performs a coarse-grained filtering by computing a unified weight map across the spatio-temporal dimensions, while the SSTA module conducts a more fine-grained refinement by separately modeling temporal and spatial attentions. Extensive experiments conducted on the CMLR and GRID datasets demonstrate that MA-LipNet significantly reduces the Character Error Rate (CER) and Word Error Rate (WER), validating its effectiveness and superiority over several state-of-the-art methods. Our work highlights the importance of multi-dimensional feature refinement for robust visual speech recognition.

Factored Reasoning with Inner Speech and Persistent Memory for Evidence-Grounded Human-Robot Interaction

Dialogue-based human-robot interaction requires robot cognitive assistants to maintain persistent user context, recover from underspecified requests, and ground responses in external evidence, while keeping intermediate decisions verifiable. In this paper we introduce JANUS, a cognitive architecture for assistive robots that models interaction as a partially observable Markov decision process and realizes control as a factored controller with typed interfaces. To this aim, Janus (i) decomposes the overall behavior into specialized modules, related to scope detection, intent recognition, memory, inner speech, query generation, and outer speech, and (ii) exposes explicit policies for information sufficiency, execution readiness, and tool grounding. A dedicated memory agent maintains a bounded recent-history buffer, a compact core memory, and an archival store with semantic retrieval, coupled through controlled consolidation and revision policies. Models inspired by the notion of inner speech in cognitive theories provide a control-oriented internal textual flow that validates parameter completeness and triggers clarification before grounding, while a faithfulness constraint ties robot-to-human claims to an evidence bundle combining working context and retrieved tool outputs. We evaluate JANUS through module-level unit tests in a dietary assistance domain grounded on a knowledge graph, reporting high agreement with curated references and practical latency profiles. These results support factored reasoning as a promising path to scalable, auditable, and evidence-grounded robot assistance over extended interaction horizons.

DementiaBank-Emotion: A Multi-Rater Emotion Annotation Corpus for Alzheimer's Disease Speech (Version 1.0)

We present DementiaBank-Emotion, the first multi-rater emotion annotation corpus for Alzheimer's disease (AD) speech. Annotating 1,492 utterances from 108 speakers for Ekman's six basic emotions and neutral, we find that AD patients express significantly more non-neutral emotions (16.9%) than healthy controls (5.7%; p < .001). Exploratory acoustic analysis suggests a possible dissociation: control speakers showed substantial F0 modulation for sadness (Delta = -3.45 semitones from baseline), whereas AD speakers showed minimal change (Delta = +0.11 semitones; interaction p = .023), though this finding is based on limited samples (sadness: n=5 control, n=15 AD) and requires replication. Within AD speech, loudness differentiates emotion categories, indicating partially preserved emotion-prosody mappings. We release the corpus, annotation guidelines, and calibration workshop materials to support research on emotion recognition in clinical populations.

A Baseline Multimodal Approach to Emotion Recognition in Conversations

We present a lightweight multimodal baseline for emotion recognition in conversations using the SemEval-2024 Task 3 dataset built from the sitcom Friends. The goal of this report is not to propose a novel state-of-the-art method, but to document an accessible reference implementation that combines (i) a transformer-based text classifier and (ii) a self-supervised speech representation model, with a simple late-fusion ensemble. We report the baseline setup and empirical results obtained under a limited training protocol, highlighting when multimodal fusion improves over unimodal models. This preprint is provided for transparency and to support future, more rigorous comparisons.

An Effective Energy Mask-based Adversarial Evasion Attacks against Misclassification in Speaker Recognition Systems

Evasion attacks pose significant threats to AI systems, exploiting vulnerabilities in machine learning models to bypass detection mechanisms. The widespread use of voice data, including deepfakes, in promising future industries is currently hindered by insufficient legal frameworks. Adversarial attack methods have emerged as the most effective countermeasure against the indiscriminate use of such data. This research introduces masked energy perturbation (MEP), a novel approach using power spectrum for energy masking of original voice data. MEP applies masking to small energy regions in the frequency domain before generating adversarial perturbations, targeting areas less noticeable to the human auditory model. The study primarily employs advanced speaker recognition models, including ECAPA-TDNN and ResNet34, which have shown remarkable performance in speaker verification tasks. The proposed MEP method demonstrated strong performance in both audio quality and evasion effectiveness. The energy masking approach effectively minimizes the perceptual evaluation of speech quality (PESQ) degradation, indicating that minimal perceptual distortion occurs to the human listener despite the adversarial perturbations. Specifically, in the PESQ evaluation, the relative performance of the MEP method was 26.68% when compared to the fast gradient sign method (FGSM) and iterative FGSM.