arXiv subjects

Yusuke Fujita

Publications and source records attributed to Yusuke Fujita.

At least 19 recordsLinked to original sources

arXiv2026

Sarashina2.2-TTS: Tackling Kanji Polyphony in Japanese Speech Generation via Data Scaling and Targeted Data Synthesis

While large language model (LLM)-based text-to-speech (TTS) systems have achieved high-quality speech synthesis, most existing systems focus on English and Chinese. Japanese, however, remains under-explored, and its unique linguistic challenges, such as widespread context-dependent kanji polyphony, have yet to be adequately tackled. Here we introduce Sarashina2.2-TTS (https://github.com/sbintuitions/sarashina2.2-tts), a Japanese-centric LLM-TTS system that tackles these challenges through a dual approach: data strategy and evaluation methodology. First, we scale training to approximately 361k hours of speech, incorporating a balanced mix of Japanese and English data. Furthermore, we design a targeted data augmentation pipeline covering all 2,136 Joyo (regular-use) kanji designated by Japan's Agency for Cultural Affairs to efficiently address kanji polyphony disambiguation. Second, we introduce the Joyo Kanji Yomi Benchmark (https://github.com/sbintuitions/JoyoKanji-Yomi-Benchmark), covering all 2,136 Joyo kanji and their 4,378 readings. Alongside this benchmark, we propose Kana-CER, a metric that compares synthesized speech against reference readings in the kana space, eliminating orthographic variations to directly measure pronunciation correctness. Experiments demonstrate that our targeted data augmentation significantly improves reading accuracy. Overall, Sarashina2.2-TTS achieves state-of-the-art kanji-level reading accuracy and matches top baselines on general sentence-level pronunciation, while delivering the highest speaker similarity in zero-shot Japanese speech synthesis. Furthermore, cross-lingual evaluation reveals that Sarashina2.2-TTS is the only system that maintains stable Japanese pronunciation regardless of the prompt language, confirming that our balanced training approach improves cross-lingual robustness.

cs.SD↗

arXiv2026

Evaluating Japanese Dialect Robustness Across Speech and Text-based Large Language Models

Dialogue systems based on large language models (LLMs) have advanced significantly in recent years. However, dialectal variation remains a major challenge, particularly for systems that process spoken input. LLM-based speech language models (SLMs), which integrate LLMs with speech processing components, show promise for spoken language tasks, yet their ability to comprehend dialects has not been sufficiently studied. Moreover, it remains unclear how the dialectal understanding of the base LLM affects SLM performance. This study investigates the dialectal robustness of both LLMs and SLMs using Japanese dialects as a test case. We define robustness as the ratio of performance on dialectal versus standard inputs, enabling fair comparisons. Our experiments show that SLM robustness correlates with that of their text-based counterparts. Furthermore, training with dialectal data and fine-tuning the speech encoder each improves robustness in SLMs.

eess.AS↗

arXiv2026

Does Translation-Enhanced Speech Encoder Pre-training Affect Speech LLMs?

Connecting a pre-trained speech encoder to a Large Language Model (LLM) is the standard architecture for building Speech LLMs. However, a structural misalignment exists between the encoder and the LLM. Unlike encoders based on automatic speech recognition, which often produce representations in separate language-specific spaces, LLMs operate within a unified language-agnostic space. A mechanism is required to align the encoder's language-specific representations with the LLM's shared space. We argue that speech translation provides a principled way to achieve this. Unlike monolingual transcription, translation requires the model to bridge different languages and learn language-agnostic representations. We experimentally evaluate the impact of incorporating translation objectives into speech encoder pre-training. Our results demonstrate that translation-enhanced pre-training improves cross-modal integration and leads to superior performance across downstream Speech LLM tasks.

eess.AS↗

arXiv2026

Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization

SpeechLLMs typically combine ASR-trained encoders with text-based LLM backbones, leading them to inherit written-style output patterns unsuitable for text-to-speech synthesis. This mismatch is particularly pronounced in Japanese, where spoken and written registers differ substantially in politeness markers, sentence-final particles, and syntactic complexity. We propose a preference-based alignment approach to adapt Japanese SpeechLLMs for speech-worthy outputs: text that is concise, conversational, and readily synthesized as natural speech. To rigorously evaluate this task, we introduce SpokenElyza, a benchmark for Japanese speech-worthiness derived from ELYZA-tasks-100 with auditory verification by native experts. Experiments show that our approach achieves substantial improvement on SpokenElyza while largely preserving performance on the original written-style evaluation. We will release SpokenElyza to support future research on Japanese spoken dialog systems.

cs.SD↗

arXiv2026

Streaming Translation and Transcription Through Speech-to-Text Causal Alignment

Simultaneous machine translation (SiMT) has traditionally relied on offline machine translation models coupled with human-engineered heuristics or learned policies. We propose Hikari, a policy-free, end-to-end model for simultaneous speech-to-text translation and streaming transcription. We also introduce Decoder Time Dilation, a mechanism that counteracts the overrepresentation of WAIT tokens in training. We present a supervised fine-tuning strategy that trains the model to recover from delays, significantly improving the quality-latency trade-off. Despite its modest size, Hikari delivers competitive translation quality at consistently low latency, comparing favorably with published IWSLT 2026 submissions up to 38x larger and with proprietary API systems across en-ja, en-de, and en-ru. We release our model weights and code to facilitate further research.

cs.CL↗

arXiv2026

Distilling LLM Semantic Priors into Encoder-Only Multi-Talker ASR with Talker-Count Routing

Large language models (LLMs) provide strong semantic priors that can improve multi-talker automatic speech recognition (MT-ASR), but using an LLM as an autoregressive decoder is computationally expensive and remains fragile under heavy overlap. In this paper, we propose an encoder-only MT-ASR framework that adapts an LLM to multi-talker conditioning and distills its semantic guidance into the encoder during training, while retaining fast CTC-style decoding at inference. Our model employs a post-encoder separator with serialized CTC to produce talker-ordered transcripts, and leverages an adapted LLM-based SOT objective as a multi-talker-aware teacher signal to explicitly regularize mixed-speech representations. To further support variable numbers of talkers, we introduce a Talker-Count Head that predicts the talker count and dynamically selects the appropriate decoding branch. Experiments on LibriMix show that the proposed encoder-only model achieves comparable performance to LLM-based systems in the two-talker condition, while delivering significant improvements in the three-talker condition with significant small RTF.

cs.SD↗

arXiv2026

DuplexCascade: Full-Duplex Speech-to-Speech Dialogue with VAD-Free Cascaded ASR-LLM-TTS Pipeline and Micro-Turn Optimization

Spoken dialog systems with cascaded ASR-LLM-TTS modules retain strong LLM intelligence, but VAD segmentation often forces half-duplex turns and brittle control. On the other hand, VAD-free end-to-end model support full-duplex interaction but is hard to maintain conversational intelligence. In this paper, we present DuplexCascade, a VAD-free cascaded streaming pipeline for full-duplex speech-to-speech dialogue. Our key idea is to convert conventional utterance-wise long turns into chunk-wise micro-turn interactions, enabling rapid bidirectional exchange while preserving the strengths of a capable text LLM. To reliably coordinate turn-taking and response timing, we introduce a set of conversational special control tokens that steer the LLM's behavior under streaming constraints. On Full-DuplexBench and VoiceBench, DuplexCascade delivers state-of-the-art full-duplex turn-taking and strong conversational intelligence among open-source speech-to-speech dialogue systems.

cs.CL↗

arXiv2025

Serialized Output Prompting for Large Language Model-based Multi-Talker Speech Recognition

Prompts are crucial for task definition and for improving the performance of large language models (LLM)-based systems. However, existing LLM-based multi-talker (MT) automatic speech recognition (ASR) systems either omit prompts or rely on simple task-definition prompts, with no prior work exploring the design of prompts to enhance performance. In this paper, we propose extracting serialized output prompts (SOP) and explicitly guiding the LLM using structured prompts to improve system performance (SOP-MT-ASR). A Separator and serialized Connectionist Temporal Classification (CTC) layers are inserted after the speech encoder to separate and extract MT content from the mixed speech encoding in a first-speaking-first-out manner. Subsequently, the SOP, which serves as a prompt for LLMs, is obtained by decoding the serialized CTC outputs using greedy search. To train the model effectively, we design a three-stage training strategy, consisting of serialized output training (SOT) fine-tuning, serialized speech information extraction, and SOP-based adaptation. Experimental results on the LibriMix dataset show that, although the LLM-based SOT model performs well in the two-talker scenario, it fails to fully leverage LLMs under more complex conditions, such as the three-talker scenario. The proposed SOP approach significantly improved performance under both two- and three-talker conditions.

cs.CL↗

arXiv2025

AC/DC: LLM-based Audio Comprehension via Dialogue Continuation

We propose an instruction-following audio comprehension model that leverages the dialogue continuation ability of large language models (LLMs). Instead of directly generating target captions in training data, the proposed method trains a model to produce responses as if the input caption triggered a dialogue. This dialogue continuation training mitigates the caption variation problem. Learning to continue a dialogue effectively captures the caption's meaning beyond its surface-level words. As a result, our model enables zero-shot instruction-following capability without multitask instruction tuning, even trained solely on audio captioning datasets. Experiments on AudioCaps, WavCaps, and Clotho datasets with AudioBench audio-scene question-answering tests demonstrate our model's ability to follow various unseen instructions.

eess.AS↗

arXiv2025

OWSM-Biasing: Contextualizing Open Whisper-Style Speech Models for Automatic Speech Recognition with Dynamic Vocabulary

Speech foundation models (SFMs), such as Open Whisper-Style Speech Models (OWSM), are trained on massive datasets to achieve accurate automatic speech recognition. However, even SFMs struggle to accurately recognize rare and unseen words. While contextual biasing (CB) is a promising approach to improve recognition of such words, most CB methods are trained from scratch, resulting in lower performance than SFMs due to the lack of pre-trained knowledge. This paper integrates an existing CB method with OWSM v3.1 while freezing its pre-trained parameters. By leveraging the knowledge embedded in SFMs, the proposed method enables effective CB while preserving the advantages of SFMs, even with a small dataset. Experimental results show that the proposed method improves the biasing word error rate (B-WER) by 11.6 points, resulting in a 0.9 point improvement in the overall WER while reducing the real-time factor by 7.5% compared to the non-biasing baseline on the LibriSpeech 100 test-clean set.

cs.SD↗

arXiv2025

DnR-nonverbal: Cinematic Audio Source Separation Dataset Containing Non-Verbal Sounds

We propose a new dataset for cinematic audio source separation (CASS) that handles non-verbal sounds. Existing CASS datasets only contain reading-style sounds as a speech stem. These datasets differ from actual movie audio, which is more likely to include acted-out voices. Consequently, models trained on conventional datasets tend to have issues where emotionally heightened voices, such as laughter and screams, are more easily separated as an effect, not speech. To address this problem, we build a new dataset, DnR-nonverbal. The proposed dataset includes non-verbal sounds like laughter and screams in the speech stem. From the experiments, we reveal the issue of non-verbal sound extraction by the current CASS model and show that our dataset can effectively address the issue in the synthetic and actual movie audio. Our dataset is available at https://zenodo.org/records/15470640.

cs.SD↗

arXiv2025

Reynolds Number Effects on Lift Enhancement Mechanisms of Dragonfly Wings: Their Effective Ranges and Determination by Local Reynolds Numbers

A corrugated structure, rather than a smooth surface, is a characteristic feature of insect wings (e.g., dragonfly wings), which enhances their aerodynamic performance at low Reynolds numbers ($Re \simeq O(10^3)$). However, the mechanisms responsible for these improvements remain largely unexplored. Previous studies have shown that a secondary vortex forms on a flat wing, opposite in sign to the leading-edge vortex (LEV). At $Re = 4000$, the lift enhancement in the corrugated wing is associated with vortex collapse and confinement within the V-shaped region, a part of corrugated structure. Conversely, when there was no lift improvement, the vortex remained intact and erupted without collapsing. In addition, the alternating vortices within the V-shaped region, comprising a negative vortex originating from the LEV and a positive vortex from the secondary vortex, induced a strong negative pressure, thereby further enhancing the lift. However, the working range of these mechanisms has yet to be investigated. In this study, lift enhancement was investigated over a broader Reynolds number range ($100 \leq Re \leq 4000$), focusing on the effective ranges. No characteristic mechanism was observed for $100 \leq Re \leq 500$. For $1000 \leq Re \leq 4000$, the alternating vortices around the V-shaped region were correlated with the improved aerodynamic performance. Furthermore, for $2000 \leq Re \leq 4000$, the secondary vortex collapse plays a major role in lift enhancement. These findings demonstrate that the lift enhancement mechanisms for corrugated wings operate within distinct working ranges depending on the Reynolds number, thereby providing insights into bioinspired aerodynamic designs.

physics.flu-dyn↗

arXiv2025

Music Tagging with Classifier Group Chains

We propose music tagging with classifier chains that model the interplay of music tags. Most conventional methods estimate multiple tags independently by treating them as multiple independent binary classification problems. This treatment overlooks the conditional dependencies among music tags, leading to suboptimal tagging performance. Unlike most music taggers, the proposed method sequentially estimates each tag based on the idea of the classifier chains. Beyond the naive classifier chains, the proposed method groups the multiple tags by category, such as genre, and performs chains by unit of groups, which we call \textit{classifier group chains}. Our method allows the modeling of the dependence between tag groups. We evaluate the effectiveness of the proposed method for music tagging performance through music tagging experiments using the MTG-Jamendo dataset. Furthermore, we investigate the effective order of chains for music tagging.

cs.SD↗

arXiv2024

Song Data Cleansing for End-to-End Neural Singer Diarization Using Neural Analysis and Synthesis Framework

We propose a data cleansing method that utilizes a neural analysis and synthesis (NANSY++) framework to train an end-to-end neural diarization model (EEND) for singer diarization. Our proposed model converts song data with choral singing which is commonly contained in popular music and unsuitable for generating a simulated dataset to the solo singing data. This cleansing is based on NANSY++, which is a framework trained to reconstruct an input non-overlapped audio signal. We exploit the pre-trained NANSY++ to convert choral singing into clean, non-overlapped audio. This cleansing process mitigates the mislabeling of choral singing to solo singing and helps the effective training of EEND models even when the majority of available song data contains choral singing sections. We experimentally evaluated the EEND model trained with a dataset using our proposed method using annotated popular duet songs. As a result, our proposed method improved 14.8 points in diarization error rate.

eess.AS↗

arXiv2024

Audio Fingerprinting with Holographic Reduced Representations

This paper proposes an audio fingerprinting model with holographic reduced representation (HRR). The proposed method reduces the number of stored fingerprints, whereas conventional neural audio fingerprinting requires many fingerprints for each audio track to achieve high accuracy and time resolution. We utilize HRR to aggregate multiple fingerprints into a composite fingerprint via circular convolution and summation, resulting in fewer fingerprints with the same dimensional space as the original. Our search method efficiently finds a combined fingerprint in which a query fingerprint exists. Using HRR's inverse operation, it can recover the relative position within a combined fingerprint, retaining the original time resolution. Experiments show that our method can reduce the number of fingerprints with modest accuracy degradation while maintaining the time resolution, outperforming simple decimation and summation-based aggregation methods.

eess.AS↗

arXiv2024

Universal Score-based Speech Enhancement with High Content Preservation

We propose UNIVERSE++, a universal speech enhancement method based on score-based diffusion and adversarial training. Specifically, we improve the existing UNIVERSE model that decouples clean speech feature extraction and diffusion. Our contributions are three-fold. First, we make several modifications to the network architecture, improving training stability and final performance. Second, we introduce an adversarial loss to promote learning high quality speech features. Third, we propose a low-rank adaptation scheme with a phoneme fidelity loss to improve content preservation in the enhanced speech. In the experiments, we train a universal enhancement model on a large scale dataset of speech degraded by noise, reverberation, and various distortions. The results on multiple public benchmark datasets demonstrate that UNIVERSE++ compares favorably to both discriminative and generative baselines for a wide range of qualitative and intelligibility metrics.

eess.AS↗

arXiv2024

Acoustic modeling for Overlapping Speech Recognition: JHU Chime-5 Challenge System

This paper summarizes our acoustic modeling efforts in the Johns Hopkins University speech recognition system for the CHiME-5 challenge to recognize highly-overlapped dinner party speech recorded by multiple microphone arrays. We explore data augmentation approaches, neural network architectures, front-end speech dereverberation, beamforming and robust i-vector extraction with comparisons of our in-house implementations and publicly available tools. We finally achieved a word error rate of 69.4% on the development set, which is a 11.7% absolute improvement over the previous baseline of 81.1%, and release this improved baseline with refined techniques/tools as an advanced CHiME-5 recipe.

eess.AS↗

arXiv2024

Keep Decoding Parallel with Effective Knowledge Distillation from Language Models to End-to-end Speech Recognisers

This study presents a novel approach for knowledge distillation (KD) from a BERT teacher model to an automatic speech recognition (ASR) model using intermediate layers. To distil the teacher's knowledge, we use an attention decoder that learns from BERT's token probabilities. Our method shows that language model (LM) information can be more effectively distilled into an ASR model using both the intermediate layers and the final layer. By using the intermediate layers as distillation target, we can more effectively distil LM knowledge into the lower network layers. Using our method, we achieve better recognition accuracy than with shallow fusion of an external LM, allowing us to maintain fast parallel decoding. Experiments on the LibriSpeech dataset demonstrate the effectiveness of our approach in enhancing greedy decoding with connectionist temporal classification (CTC).

cs.CL↗