Tomoki Toda

Lester Phillip Violeta, Xueyao Zhang, Jiatong Shi et al.

We present a thorough analysis of the findings of the latest iteration of the Singing Voice Conversion Challenge, a scientific event aiming to compare and understand different voice conversion systems in a controlled environment. Compared to previous iterations which solely focused on converting the singer identity, this year we also focused on converting the singing style of the singer. To create a controlled environment and thorough evaluations, we developed a new challenge database, introduced two tasks, open-sourced baselines, and conducted large-scale crowd-sourced listening tests and objective evaluations. The challenge was run for two months and in total we evaluated 33 different systems. The results of the large-scale crowd-sourced listening test showed that top systems had comparable singer identity scores to ground truth samples. However, modeling the singing style and consequently achieving high naturalness still remains a challenge in this task, primarily due to the difficulty in modeling dynamic information in breathy, glissando, and vibrato singing styles. Further analyses of the challenge also discuss the limitations of both the traditional similarity test and the dynamic preference test in evaluating singing style similarity. Moreover, calculating Spearman's rank correlation coefficient shows that dependent objective metrics such as chroma-alignment and non-match metrics such as speaker embeddings are the most correlated to subjective scores, but are still not at a level where it could be considered as a true replacement for subjective scores.

Wen-Chin Huang, Shu-Wen Yang, Tomoki Hayashi et al.

We present a large-scale comparative study of self-supervised speech representation (S3R)-based voice conversion (VC). In the context of recognition-synthesis VC, S3Rs are attractive owing to their potential to replace expensive supervised representations such as phonetic posteriorgrams (PPGs), which are commonly adopted by state-of-the-art VC systems. Using S3PRL-VC, an open-source VC software we previously developed, we provide a series of in-depth objective and subjective analyses under three VC settings: intra-/cross-lingual any-to-one (A2O) and any-to-any (A2A) VC, using the voice conversion challenge 2020 (VCC2020) dataset. We investigated S3R-based VC in various aspects, including model type, multilinguality, and supervision. We also studied the effect of a post-discretization process with k-means clustering and showed how it improves in the A2A setting. Finally, the comparison with state-of-the-art VC systems demonstrates the competitiveness of S3R-based VC and also sheds light on the possible improving directions.

Ryuichi Yamamoto, Reo Yoneyama, Tomoki Toda

This paper describes the design of NNSVS, an open-source software for neural network-based singing voice synthesis research. NNSVS is inspired by Sinsy, an open-source pioneer in singing voice synthesis research, and provides many additional features such as multi-stream models, autoregressive fundamental frequency models, and neural vocoders. Furthermore, NNSVS provides extensive documentation and numerous scripts to build complete singing voice synthesis systems. Experimental results demonstrate that our best system significantly outperforms our reproduction of Sinsy and other baseline systems. The toolkit is available at https://github.com/nnsvs/nnsvs.

Wen-Chin Huang, Tomoki Toda

Foreign accent conversion (FAC) is a special application of voice conversion (VC) which aims to convert the accented speech of a non-native speaker to a native-sounding speech with the same speaker identity. FAC is difficult since the native speech from the desired non-native speaker to be used as the training target is impossible to collect. In this work, we evaluate three recently proposed methods for ground-truth-free FAC, where all of them aim to harness the power of sequence-to-sequence (seq2seq) and non-parallel VC models to properly convert the accent and control the speaker identity. Our experimental evaluation results show that no single method was significantly better than the others in all evaluation axes, which is in contrast to conclusions drawn in previous studies. We also explain the effectiveness of these methods with the training input and output of the seq2seq model and examine the design choice of the non-parallel VC model, and show that intelligibility measures such as word error rates do not correlate well with subjective accentedness. Finally, our implementation is open-sourced to promote reproducible research and help future researchers improve upon the compared systems.

Jinyi Mi, Ding Ma, Tomoki Toda

Zero-shot cross-lingual speech emotion recognition (SER) remains challenging due to distribution mismatches across languages and the lack of emotion annotations in target language. Under such conditions, models trained solely on source-language data frequently suffer from degraded generalization when evaluated on unseen target languages. To address this limitation, we propose an emotion-discriminative representation learning method that integrates supervised contrastive learning and speaker adversarial learning. The contrastive learning promotes cross-lingual emotion alignment, while speaker adversarial learning suppresses speaker-related cues to encourage speaker-invariant representations. Experimental results under a zero-shot cross-lingual SER setting demonstrate that the proposed method significantly improves SER performance over conventional training strategies.

Wen-Chin Huang, Lester Phillip Violeta, Songxiang Liu et al.

We present the latest iteration of the voice conversion challenge (VCC) series, a bi-annual scientific event aiming to compare and understand different voice conversion (VC) systems based on a common dataset. This year we shifted our focus to singing voice conversion (SVC), thus named the challenge the Singing Voice Conversion Challenge (SVCC). A new database was constructed for two tasks, namely in-domain and cross-domain SVC. The challenge was run for two months, and in total we received 26 submissions, including 2 baselines. Through a large-scale crowd-sourced listening test, we observed that for both tasks, although human-level naturalness was achieved by the top system, no team was able to obtain a similarity score as high as the target speakers. Also, as expected, cross-domain SVC is harder than in-domain SVC, especially in the similarity aspect. We also investigated whether existing objective measurements were able to predict perceptual performance, and found that only few of them could reach a significant correlation.

Ding Ma, Jinyi Mi, Fengji Li et al.

Objective: laryngectomees depend on an electromechanical device to generate electrolaryngeal (EL) speech. Compared with normal speech, EL speech suffers from severe distortion, limited phonetic variation, unnatural prosody, and temporal shifts, degrading naturalness and intelligibility. Although sequence-to-sequence (seq2seq) voice conversion (VC) based EL-speech-to-normal-speech conversion (EL2SP) is promising, substantial mismatches between EL and normal speech inevitably cause cumulative mapping errors that limit performance. To address this, we describe a novel representation learning framework integrating speech and text representations to improve mapping and reconstruction quality within a seq2seq VC model. Methods: our methodology comprises two main stages: 1) representation integration and learning, and 2) reconstruction training. A network capable of incorporating auxiliary text information is first constructed with pretrained modules to learn speech--text-based integrated representations. Then, an autoencoder-style reconstruction strategy finalizes EL2SP model to inherit these representations without increasing model complexity. We introduce three fusion strategies including middle-, input-, and hybrid-level fusion strategies that progressively enhance learning. Moreover, besides standard seq2seq VC objectives, an additional reconstruction loss on the integrated representation is introduced to refine representation transfer. Results: experiments under different EL2SP datasets consistently demonstrate that our methods, combined with data augmentations, outperform baselines relying solely on speech representations. Furthermore, progressive improvements with system design depth validate the effectiveness of our methods. Significance: the proposed methods provide an extensible and practical methodology for EL speech enhancement and assistive communication technologies.

Reo Yoneyama, Yi-Chiao Wu, Tomoki Toda

Our previous work, the unified source-filter GAN (uSFGAN) vocoder, introduced a novel architecture based on the source-filter theory into the parallel waveform generative adversarial network to achieve high voice quality and pitch controllability. However, the high temporal resolution inputs result in high computation costs. Although the HiFi-GAN vocoder achieves fast high-fidelity voice generation thanks to the efficient upsampling-based generator architecture, the pitch controllability is severely limited. To realize a fast and pitch-controllable high-fidelity neural vocoder, we introduce the source-filter theory into HiFi-GAN by hierarchically conditioning the resonance filtering network on a well-estimated source excitation information. According to the experimental results, our proposed method outperforms HiFi-GAN and uSFGAN on a singing voice generation in voice quality and synthesis speed on a single CPU. Furthermore, unlike the uSFGAN vocoder, the proposed method can be easily adopted/integrated in real-time applications and end-to-end systems.

Wen-Chin Huang, Erica Cooper, Tomoki Toda

In this paper, we study the task of subjective speech quality assessment (SSQA), which refers to predicting the perceptual quality of speech. Owing to the development of deep neural network models, SSQA has greatly advanced and has been widely applied in scientific papers to evaluate speech generation systems. Nonetheless, the insufficient out-of-domain (OOD) generalization ability of current SSQA models is underexplored and often overlooked by researchers. To study this problem systematically, we present MOS-Bench, a diverse SSQA dataset collection that currently contains 8 training sets and 17 test sets. Through extensive experiments, we first highlight the OOD generalization challenges of existing models. We then evaluate the efficacy of multiple-dataset training, comparing straightforward data pooling against AlignNet, an existing domain-aware method. We demonstrate that pooling multiple training sets provides a simple yet effective solution, and variation in the data is a key factor for robust generalization beyond training data size.

Reo Yoneyama, Yi-Chiao Wu, Tomoki Toda

This paper introduces a unified source-filter network with a harmonic-plus-noise source excitation generation mechanism. In our previous work, we proposed unified Source-Filter GAN (uSFGAN) for developing a high-fidelity neural vocoder with flexible voice controllability using a unified source-filter neural network architecture. However, the capability of uSFGAN to model the aperiodic source excitation signal is insufficient, and there is still a gap in sound quality between the natural and generated speech. To improve the source excitation modeling and generated sound quality, a new source excitation generation network separately generating periodic and aperiodic components is proposed. The advanced adversarial training procedure of HiFiGAN is also adopted to replace that of Parallel WaveGAN used in the original uSFGAN. Both objective and subjective evaluation results show that the modified uSFGAN significantly improves the sound quality of the basic uSFGAN while maintaining the voice controllability.

Yusuke Yasuda, Tomoki Toda

End-to-end text-to-speech synthesis (TTS) can generate highly natural synthetic speech from raw text. However, rendering the correct pitch accents is still a challenging problem for end-to-end TTS. To tackle the challenge of rendering correct pitch accent in Japanese end-to-end TTS, we adopt PnG~BERT, a self-supervised pretrained model in the character and phoneme domain for TTS. We investigate the effects of features captured by PnG~BERT on Japanese TTS by modifying the fine-tuning condition to determine the conditions helpful inferring pitch accents. We manipulate content of PnG~BERT features from being text-oriented to speech-oriented by changing the number of fine-tuned layers during TTS. In addition, we teach PnG~BERT pitch accent information by fine-tuning with tone prediction as an additional downstream task. Our experimental results show that the features of PnG~BERT captured by pretraining contain information helpful inferring pitch accent, and PnG~BERT outperforms baseline Tacotron on accent correctness in a listening test.

Ryuichi Yamamoto, Reo Yoneyama, Lester Phillip Violeta et al.

This paper presents our systems (denoted as T13) for the singing voice conversion challenge (SVCC) 2023. For both in-domain and cross-domain English singing voice conversion (SVC) tasks (Task 1 and Task 2), we adopt a recognition-synthesis approach with self-supervised learning-based representation. To achieve data-efficient SVC with a limited amount of target singer/speaker's data (150 to 160 utterances for SVCC 2023), we first train a diffusion-based any-to-any voice conversion model using publicly available large-scale 750 hours of speech and singing data. Then, we finetune the model for each target singer/speaker of Task 1 and Task 2. Large-scale listening tests conducted by SVCC 2023 show that our T13 system achieves competitive naturalness and speaker similarity for the harder cross-domain SVC (Task 2), which implies the generalization ability of our proposed method. Our objective evaluation results show that using large datasets is particularly beneficial for cross-domain SVC.

Jiajun He, Zekun Yang, Tomoki Toda

Automatic speech recognition (ASR) systems often encounter difficulties in accurately recognizing rare words, leading to errors that can have a negative impact on downstream tasks such as keyword spotting, intent detection, and text summarization. To address this challenge, we present a novel ASR postprocessing method that focuses on improving the recognition of rare words through error detection and context-aware error correction. Our method optimizes the decoding process by targeting only the predicted error positions, minimizing unnecessary computations. Moreover, we leverage a rare word list to provide additional contextual knowledge, enabling the model to better correct rare words. Experimental results across five datasets demonstrate that our proposed method achieves significantly lower word error rates (WERs) than previous approaches while maintaining a reasonable inference speed. Furthermore, our approach exhibits promising robustness across different ASR systems.

Yusuke Yasuda, Tomoki Toda

Text-to-speech synthesis (TTS) is a task to convert texts into speech. Two of the factors that have been driving TTS are the advancements of probabilistic models and latent representation learning. We propose a TTS method based on latent variable conversion using a diffusion probabilistic model and the variational autoencoder (VAE). In our TTS method, we use a waveform model based on VAE, a diffusion model that predicts the distribution of latent variables in the waveform model from texts, and an alignment model that learns alignments between the text and speech latent sequences. Our method integrates diffusion with VAE by modeling both mean and variance parameters with diffusion, where the target distribution is determined by approximation from VAE. This latent variable conversion framework potentially enables us to flexibly incorporate various latent feature extractors. Our experiments show that our method is robust to linguistic labels with poor orthography and alignment errors.

Tatsuya Komatsu, Yusuke Fujita, Kazuya Takeda et al.

This study introduces a novel training paradigm, audio difference learning, for improving audio captioning. The fundamental concept of the proposed learning method is to create a feature representation space that preserves the relationship between audio, enabling the generation of captions that detail intricate audio information. This method employs a reference audio along with the input audio, both of which are transformed into feature representations via a shared encoder. Captions are then generated from these differential features to describe their differences. Furthermore, a unique technique is proposed that involves mixing the input audio with additional audio, and using the additional audio as a reference. This results in the difference between the mixed audio and the reference audio reverting back to the original input audio. This allows the original input's caption to be used as the caption for their difference, eliminating the need for additional annotations for the differences. In the experiments using the Clotho and ESC50 datasets, the proposed method demonstrated an improvement in the SPIDEr score by 7% compared to conventional methods.

Xiaohan Shi, Jiajun He, Xingfeng Li et al.

This paper proposes an efficient attempt to noisy speech emotion recognition (NSER). Conventional NSER approaches have proven effective in mitigating the impact of artificial noise sources, such as white Gaussian noise, but are limited to non-stationary noises in real-world environments due to their complexity and uncertainty. To overcome this limitation, we introduce a new method for NSER by adopting the automatic speech recognition (ASR) model as a noise-robust feature extractor to eliminate non-vocal information in noisy speech. We first obtain intermediate layer information from the ASR model as a feature representation for emotional speech and then apply this representation for the downstream NSER task. Our experimental results show that 1) the proposed method achieves better NSER performance compared with the conventional noise reduction method, 2) outperforms self-supervised learning approaches, and 3) even outperforms text-based approaches using ASR transcription or the ground truth transcription of noisy speech.

Jinyi Mi, Xiaohan Shi, Ding Ma et al.

Developing a robust speech emotion recognition (SER) system in noisy conditions faces challenges posed by different noise properties. Most previous studies have not considered the impact of human speech noise, thus limiting the application scope of SER. In this paper, we propose a novel two-stage framework for the problem by cascading target speaker extraction (TSE) method and SER. We first train a TSE model to extract the speech of target speaker from a mixture. Then, in the second stage, we utilize the extracted speech for SER training. Additionally, we explore a joint training of TSE and SER models in the second stage. Our developed system achieves a 14.33% improvement in unweighted accuracy (UA) compared to a baseline without using TSE method, demonstrating the effectiveness of our framework in mitigating the impact of human speech noise. Moreover, we conduct experiments considering speaker gender, showing that our framework performs particularly well in different-gender mixture.

Xingfeng Li, Xiaohan Shi, Junjie Li et al.

This study introduces EM2LDL, a novel multilingual speech corpus designed to advance mixed emotion recognition through label distribution learning. Addressing the limitations of predominantly monolingual and single-label emotion corpora \textcolor{black}{that restrict linguistic diversity, are unable to model mixed emotions, and lack ecological validity}, EM2LDL comprises expressive utterances in English, Mandarin, and Cantonese, capturing the intra-utterance code-switching prevalent in multilingual regions like Hong Kong and Macao. The corpus integrates spontaneous emotional expressions from online platforms, annotated with fine-grained emotion distributions across 32 categories. Experimental baselines using self-supervised learning models demonstrate robust performance in speaker-independent gender-, age-, and personality-based evaluations, with HuBERT-large-EN achieving optimal results. By incorporating linguistic diversity and ecological validity, EM2LDL enables the exploration of complex emotional dynamics in multilingual settings. This work provides a versatile testbed for developing adaptive, empathetic systems for applications in affective computing, including mental health monitoring and cross-cultural communication. The dataset, annotations, and baseline codes are publicly available at https://github.com/xingfengli/EM2LDL.

Wen-Chin Huang, Shu-Wen Yang, Tomoki Hayashi et al.

This paper introduces S3PRL-VC, an open-source voice conversion (VC) framework based on the S3PRL toolkit. In the context of recognition-synthesis VC, self-supervised speech representation (S3R) is valuable in its potential to replace the expensive supervised representation adopted by state-of-the-art VC systems. Moreover, we claim that VC is a good probing task for S3R analysis. In this work, we provide a series of in-depth analyses by benchmarking on the two tasks in VCC2020, namely intra-/cross-lingual any-to-one (A2O) VC, as well as an any-to-any (A2A) setting. We also provide comparisons between not only different S3Rs but also top systems in VCC2020 with supervised representations. Systematic objective and subjective evaluation were conducted, and we show that S3R is comparable with VCC2020 top systems in the A2O setting in terms of similarity, and achieves state-of-the-art in S3R-based A2A VC. We believe the extensive analysis, as well as the toolkit itself, contribute to not only the S3R community but also the VC community. The codebase is now open-sourced.

Kazuhiro Kobayashi, Wen-Chin Huang, Yi-Chiao Wu et al.

In this paper, we present an open-source software for developing a nonparallel voice conversion (VC) system named crank. Although we have released an open-source VC software based on the Gaussian mixture model named sprocket in the last VC Challenge, it is not straightforward to apply any speech corpus because it is necessary to prepare parallel utterances of source and target speakers to model a statistical conversion function. To address this issue, in this study, we developed a new open-source VC software that enables users to model the conversion function by using only a nonparallel speech corpus. For implementing the VC software, we used a vector-quantized variational autoencoder (VQVAE). To rapidly examine the effectiveness of recent technologies developed in this research field, crank also supports several representative works for autoencoder-based VC methods such as the use of hierarchical architectures, cyclic architectures, generative adversarial networks, speaker adversarial training, and neural vocoders. Moreover, it is possible to automatically estimate objective measures such as mel-cepstrum distortion and pseudo mean opinion score based on MOSNet. In this paper, we describe representative functions developed in crank and make brief comparisons by objective evaluations.

Patrick Lumban Tobing, Yi-Chiao Wu, Tomoki Toda

In this paper, we present a description of the baseline system of Voice Conversion Challenge (VCC) 2020 with a cyclic variational autoencoder (CycleVAE) and Parallel WaveGAN (PWG), i.e., CycleVAEPWG. CycleVAE is a nonparallel VAE-based voice conversion that utilizes converted acoustic features to consider cyclically reconstructed spectra during optimization. On the other hand, PWG is a non-autoregressive neural vocoder that is based on a generative adversarial network for a high-quality and fast waveform generator. In practice, the CycleVAEPWG system can be straightforwardly developed with the VCC 2020 dataset using a unified model for both Task 1 (intralingual) and Task 2 (cross-lingual), where our open-source implementation is available at https://github.com/bigpon/vcc20_baseline_cyclevae. The results of VCC 2020 have demonstrated that the CycleVAEPWG baseline achieves the following: 1) a mean opinion score (MOS) of 2.87 in naturalness and a speaker similarity percentage (Sim) of 75.37% for Task 1, and 2) a MOS of 2.56 and a Sim of 56.46% for Task 2, showing an approximately or nearly average score for naturalness and an above average score for speaker similarity.

Wen-Chin Huang, Tomoki Hayashi, Shinji Watanabe et al.

This paper presents the sequence-to-sequence (seq2seq) baseline system for the voice conversion challenge (VCC) 2020. We consider a naive approach for voice conversion (VC), which is to first transcribe the input speech with an automatic speech recognition (ASR) model, followed using the transcriptions to generate the voice of the target with a text-to-speech (TTS) model. We revisit this method under a sequence-to-sequence (seq2seq) framework by utilizing ESPnet, an open-source end-to-end speech processing toolkit, and the many well-configured pretrained models provided by the community. Official evaluation results show that our system comes out top among the participating systems in terms of conversion similarity, demonstrating the promising ability of seq2seq models to convert speaker identity. The implementation is made open-source at: https://github.com/espnet/espnet/tree/master/egs/vcc20.

Tomoki Hayashi, Ryuichi Yamamoto, Katsuki Inoue et al.

This paper introduces a new end-to-end text-to-speech (E2E-TTS) toolkit named ESPnet-TTS, which is an extension of the open-source speech processing toolkit ESPnet. The toolkit supports state-of-the-art E2E-TTS models, including Tacotron~2, Transformer TTS, and FastSpeech, and also provides recipes inspired by the Kaldi automatic speech recognition (ASR) toolkit. The recipes are based on the design unified with the ESPnet ASR recipe, providing high reproducibility. The toolkit also provides pre-trained models and samples of all of the recipes so that users can use it as a baseline. Furthermore, the unified design enables the integration of ASR functions with TTS, e.g., ASR-based objective evaluation and semi-supervised learning with both ASR and TTS models. This paper describes the design of the toolkit and experimental evaluation in comparison with other toolkits. The experimental results show that our models can achieve state-of-the-art performance comparable to the other latest toolkits, resulting in a mean opinion score (MOS) of 4.25 on the LJSpeech dataset. The toolkit is publicly available at https://github.com/espnet/espnet.

Hideki Kawahara, Ken-Ichi Sakakibara, Masanori Morise et al.

We introduce a simple and linear SNR (strictly speaking, periodic to random power ratio) estimator (0dB to 80dB without additional calibration/linearization) for providing reliable descriptions of aperiodicity in speech corpus. The main idea of this method is to estimate the background random noise level without directly extracting the background noise. The proposed method is applicable to a wide variety of time windowing functions with very low sidelobe levels. The estimate combines the frequency derivative and the time-frequency derivative of the mapping from filter center frequency to the output instantaneous frequency. This procedure can replace the periodicity detection and aperiodicity estimation subsystems of recently introduced open source vocoder, YANG vocoder. Source code of MATLAB implementation of this method will also be open sourced.

Hideki Kawahara, Ken-Ichi Sakakibara, Hideki Banno et al.

We formulated and implemented a procedure to generate aliasing-free excitation source signals. It uses a new antialiasing filter in the continuous time domain followed by an IIR digital filter for response equalization. We introduced a cosine-series-based general design procedure for the new antialiasing function. We applied this new procedure to implement the antialiased Fujisaki-Ljungqvist model. We also applied it to revise our previous implementation of the antialiased Fant-Liljencrants model. A combination of these signals and a lattice implementation of the time varying vocal tract model provides a reliable and flexible basis to test fo extractors and source aperiodicity analysis methods. MATLAB implementations of these antialiased excitation source models are available as part of our open source tools for speech science.

You Zhang, Yongyi Zang, Jiatong Shi et al.

The rapid advancement of AI-generated singing voices, which now closely mimic natural human singing and align seamlessly with musical scores, has led to heightened concerns for artists and the music industry. Unlike spoken voice, singing voice presents unique challenges due to its musical nature and the presence of strong background music, making singing voice deepfake detection (SVDD) a specialized field requiring focused attention. To promote SVDD research, we recently proposed the "SVDD Challenge," the very first research challenge focusing on SVDD for lab-controlled and in-the-wild bonafide and deepfake singing voice recordings. The challenge will be held in conjunction with the 2024 IEEE Spoken Language Technology Workshop (SLT 2024).

Jiajun He, Tomoki Toda

End-to-end automatic speech recognition (ASR) models often struggle to accurately recognize rare words. Previously, we introduced an ASR postprocessing method called error detection and context-aware error correction (ED-CEC), which leverages contextual information such as named entities and technical terms to improve the accuracy of ASR transcripts. Although ED-CEC achieves a notable success in correcting rare words, its accuracy remains low when dealing with rare words that have similar pronunciations but different spellings. To address this issue, we proposed a phoneme-augmented multimodal fusion method for context-aware error correction (PMF-CEC) method on the basis of ED-CEC, which allowed for better differentiation between target rare words and homophones. Additionally, we observed that the previous ASR error detection module suffers from overdetection. To mitigate this, we introduced a retention probability mechanism to filter out editing operations with confidence scores below a set threshold, preserving the original operation to improve error detection accuracy. Experiments conducted on five datasets demonstrated that our proposed PMF-CEC maintains reasonable inference speed while further reducing the biased word error rate compared with ED-CEC, showing a stronger advantage in correcting homophones. Moreover, our method outperforms other contextual biasing methods, and remains valuable compared with LLM-based methods in terms of faster inference and better robustness under large biasing lists.

Jiajun He, Jinyi Mi, Tomoki Toda

Multimodal emotion recognition (MER) extracts emotions from multimodal data, including visual, speech, and text inputs, playing a key role in human-computer interaction. Attention-based fusion methods dominate MER research, achieving strong classification performance. However, two key challenges remain: effectively extracting modality-specific features and capturing cross-modal similarities despite distribution differences caused by modality heterogeneity. To address these, we propose a gated interactive attention mechanism to adaptively extract modality-specific features while enhancing emotional information through pairwise interactions. Additionally, we introduce a modality-invariant generator to learn modality-invariant representations and constrain domain shifts by aligning cross-modal similarities. Experiments on IEMOCAP demonstrate that our method outperforms state-of-the-art MER approaches, achieving WA 80.7% and UA 81.3%.

Yusuke Yasuda, Tomoki Toda

A preference-based subjective evaluation is a key method for evaluating generative media reliably. However, its huge combinations of pairs prohibit it from being applied to large-scale evaluation using crowdsourcing. To address this issue, we propose an automatic optimization method for preference-based subjective evaluation in terms of pair combination selections and allocation of evaluation volumes with online learning in a crowdsourcing environment. We use a preference-based online learning method based on a sorting algorithm to identify the total order of evaluation targets with minimum sample volumes. Our online learning algorithm supports parallel and asynchronous execution under fixed-budget conditions required for crowdsourcing. Our experiment on preference-based subjective evaluation of synthetic speech shows that our method successfully optimizes the test by reducing pair combinations from 351 to 83 and allocating optimal evaluation volumes for each pair ranging from 30 to 663 without compromising evaluation accuracies and wasting budget allocations.

Bence Mark Halpern, Thomas Tienkamp, Defne Abur et al.

Automatic speech intelligibility assessment is crucial for monitoring speech disorders and therapy efficacy. However, existing methods are difficult to compare: research is fragmented across private datasets with inconsistent protocols. We introduce PathBench, a unified benchmark for pathological speech assessment using public datasets. We compare reference-free, reference-text, and reference-audio methods across three protocols (Matched Content, Extended, and Full) representing how a linguist (controlled stimuli) versus machine learning specialist (maximum data) would approach the same data. We establish benchmark baselines across six datasets, enabling systematic evaluation of future methodological advances, and introduce Dual-ASR Articulatory Precision (DArtP), achieving the highest average correlation among reference-free methods.

Jiajun He, Naoki Sawada, Koichi Miyazaki et al.

Automatic speech recognition (ASR) systems struggle with domain-specific named entities, especially homophones. Contextual ASR improves recognition but often fails to capture fine-grained phoneme variations due to limited entity diversity. Moreover, prior methods treat entities as independent tokens, leading to incomplete multi-token biasing. To address these issues, we propose Phoneme-Augmented Robust Contextual ASR via COntrastive entity disambiguation (PARCO), which integrates phoneme-aware encoding, contrastive entity disambiguation, entity-level supervision, and hierarchical entity filtering. These components enhance phonetic discrimination, ensure complete entity retrieval, and reduce false positives under uncertainty. Experiments show that PARCO achieves CER of 4.22% on Chinese AISHELL-1 and WER of 11.14% on English DATA2 under 1,000 distractors, significantly outperforming baselines. PARCO also demonstrates robust gains on out-of-domain datasets like THCHS-30 and LibriSpeech.

Jiajun He, Naoki Sawada, Koichi Miyazaki et al.

In real-world applications, automatic speech recognition (ASR) systems must handle overlapping speech from multiple speakers and recognize rare words like technical terms. Traditional methods address multi-talker ASR and contextual biasing separately, limiting performance in complex scenarios. We propose a unified framework that combines multi-talker overlapping speech recognition and contextual biasing into a single task. Our ASR method integrates pretrained speech encoders and large language models (LLMs), using optimized finetuning strategies. We also introduce a two-stage filtering algorithm to efficiently identify relevant rare words from large biasing lists and incorporate them into the LLM's prompt input, enhancing rare word recognition. Experiments show that our approach outperforms traditional contextual biasing methods, achieving a WER of 7.9% on LibriMix and 32.9% on AMI SDM when the biasing size is 1,000, demonstrating its effectiveness in complex speech scenarios.

Jiajun He, Tomoki Toda

Moment retrieval aims to locate the most relevant moment in an untrimmed video based on a given natural language query. Existing solutions can be roughly categorized into moment-based and clip-based methods. The former often involves heavy computations, while the latter, due to overlooking coarse-grained information, typically underperforms compared to moment-based models. Hence, this paper proposes a novel 2-Dimensional Pointer-based Machine Reading Comprehension for Moment Retrieval Choice (2DP-2MRC) model to address the issue of imprecise localization in clip-based methods while maintaining lower computational complexity than moment-based methods. Specifically, we introduce an AV-Encoder to capture coarse-grained information at moment and video levels. Additionally, a 2D pointer encoder module is introduced to further enhance boundary detection for target moment. Extensive experiments on the HiREST dataset demonstrate that 2DP-2MRC significantly outperforms existing baseline models.

Jiajun He, Xiaohan Shi, Xingfeng Li et al.

The prevalent approach in speech emotion recognition (SER) involves integrating both audio and textual information to comprehensively identify the speaker's emotion, with the text generally obtained through automatic speech recognition (ASR). An essential issue of this approach is that ASR errors from the text modality can worsen the performance of SER. Previous studies have proposed using an auxiliary ASR error detection task to adaptively assign weights of each word in ASR hypotheses. However, this approach has limited improvement potential because it does not address the coherence of semantic information in the text. Additionally, the inherent heterogeneity of different modalities leads to distribution gaps between their representations, making their fusion challenging. Therefore, in this paper, we incorporate two auxiliary tasks, ASR error detection (AED) and ASR error correction (AEC), to enhance the semantic coherence of ASR text, and further introduce a novel multi-modal fusion (MF) method to learn shared representations across modalities. We refer to our method as MF-AED-AEC. Experimental results indicate that MF-AED-AEC significantly outperforms the baseline model by a margin of 4.1\%.

Chao Xie, Yi-Chiao Wu, Patrick Lumban Tobing et al.

Beyond the conventional voice conversion (VC) where the speaker information is converted without altering the linguistic content, the background sounds are informative and need to be retained in some real-world scenarios, such as VC in movie/video and VC in music where the voice is entangled with background sounds. As a new VC framework, we have developed a noisy-to-noisy (N2N) VC framework to convert the speaker's identity while preserving the background sounds. Although our framework consisting of a denoising module and a VC module well handles the background sounds, the VC module is sensitive to the distortion caused by the denoising module. To address this distortion issue, in this paper we propose the improved VC module to directly model the noisy speech waveform while controlling the background sounds. The experimental results have demonstrated that our improved framework significantly outperforms the previous one and achieves an acceptable score in terms of naturalness, while reaching comparable similarity performance to the upper bound of our framework.

Hsin-Tien Chiang, Yi-Chiao Wu, Cheng Yu et al.

Without the need of a clean reference, non-intrusive speech assessment methods have caught great attention for objective evaluations. Recently, deep neural network (DNN) models have been applied to build non-intrusive speech assessment approaches and confirmed to provide promising performance. However, most DNN-based approaches are designed for normal-hearing listeners without considering hearing-loss factors. In this study, we propose a DNN-based hearing aid speech assessment network (HASA-Net), formed by a bidirectional long short-term memory (BLSTM) model, to predict speech quality and intelligibility scores simultaneously according to input speech signals and specified hearing-loss patterns. To the best of our knowledge, HASA-Net is the first work to incorporate quality and intelligibility assessments utilizing a unified DNN-based non-intrusive model for hearing aids. Experimental results show that the predicted speech quality and intelligibility scores of HASA-Net are highly correlated to two well-known intrusive hearing-aid evaluation metrics, hearing aid speech quality index (HASQI) and hearing aid speech perception index (HASPI), respectively.

Wen-Chin Huang, Erica Cooper, Junichi Yamagishi et al.

An effective approach to automatically predict the subjective rating for synthetic speech is to train on a listening test dataset with human-annotated scores. Although each speech sample in the dataset is rated by several listeners, most previous works only used the mean score as the training target. In this work, we present LDNet, a unified framework for mean opinion score (MOS) prediction that predicts the listener-wise perceived quality given the input speech and the listener identity. We reflect recent advances in LD modeling, including design choices of the model architecture, and propose two inference methods that provide more stable results and efficient computation. We conduct systematic experiments on the voice conversion challenge (VCC) 2018 benchmark and a newly collected large-scale MOS dataset, providing an in-depth analysis of the proposed framework. Results show that the mean listener inference method is a better way to utilize the mean scores, whose effectiveness is more obvious when having more ratings per sample.

Wen-Chin Huang, Bence Mark Halpern, Lester Phillip Violeta et al.

We present a voice conversion framework that converts normal speech into dysarthric speech while preserving the speaker identity. Such a framework is essential for (1) clinical decision making processes and alleviation of patient stress, (2) data augmentation for dysarthric speech recognition. This is an especially challenging task since the converted samples should capture the severity of dysarthric speech while being highly natural and possessing the speaker identity of the normal speaker. To this end, we adopted a two-stage framework, which consists of a sequence-to-sequence model and a nonparallel frame-wise model. Objective and subjective evaluations were conducted on the UASpeech dataset, and results showed that the method was able to yield reasonable naturalness and capture severity aspects of the pathological speech. On the other hand, the similarity to the normal source speaker's voice was limited and requires further improvements.

Chao Xie, Yi-Chiao Wu, Patrick Lumban Tobing et al.

In a conventional voice conversion (VC) framework, a VC model is often trained with a clean dataset consisting of speech data carefully recorded and selected by minimizing background interference. However, collecting such a high-quality dataset is expensive and time-consuming. Leveraging crowd-sourced speech data in training is more economical. Moreover, for some real-world VC scenarios such as VC in video and VC-based data augmentation for speech recognition systems, the background sounds themselves are also informative and need to be maintained. In this paper, to explore VC with the flexibility of handling background sounds, we propose a noisy-to-noisy (N2N) VC framework composed of a denoising module and a VC module. With the proposed framework, we can convert the speaker's identity while preserving the background sounds. Both objective and subjective evaluations are conducted, and the results reveal the effectiveness of the proposed framework.

Yi-Syuan Liou, Wen-Chin Huang, Ming-Chi Yen et al.

Voice conversion (VC) is an effective approach to electrolaryngeal (EL) speech enhancement, a task that aims to improve the quality of the artificial voice from an electrolarynx device. In frame-based VC methods, time alignment needs to be performed prior to model training, and the dynamic time warping (DTW) algorithm is widely adopted to compute the best time alignment between each utterance pair. The validity is based on the assumption that the same phonemes of the speakers have similar features and can be mapped by measuring a pre-defined distance between speech frames of the source and the target. However, the special characteristics of the EL speech can break the assumption, resulting in a sub-optimal DTW alignment. In this work, we propose to use lip images for time alignment, as we assume that the lip movements of laryngectomee remain normal compared to healthy people. We investigate two naive lip representations and distance metrics, and experimental results demonstrate that the proposed method can significantly outperform the audio-only alignment in terms of objective and subjective evaluations.

Wen-Chin Huang, Tomoki Hayashi, Xinjian Li et al.

In voice conversion (VC), an approach showing promising results in the latest voice conversion challenge (VCC) 2020 is to first use an automatic speech recognition (ASR) model to transcribe the source speech into the underlying linguistic contents; these are then used as input by a text-to-speech (TTS) system to generate the converted speech. Such a paradigm, referred to as ASR+TTS, overlooks the modeling of prosody, which plays an important role in speech naturalness and conversion similarity. Although some researchers have considered transferring prosodic clues from the source speech, there arises a speaker mismatch during training and conversion. To address this issue, in this work, we propose to directly predict prosody from the linguistic representation in a target-speaker-dependent manner, referred to as target text prediction (TTP). We evaluate both methods on the VCC2020 benchmark and consider different linguistic representations. The results demonstrate the effectiveness of TTP in both objective and subjective evaluations.

Ibuki Kuroyanagi, Tomoki Hayashi, Kazuya Takeda et al.

An anomalous sound detection system to detect unknown anomalous sounds usually needs to be built using only normal sound data. Moreover, it is desirable to improve the system by effectively using a small amount of anomalous sound data, which will be accumulated through the system's operation. As one of the methods to meet these requirements, we focus on a binary classification model that is developed by using not only normal data but also outlier data in the other domains as pseudo-anomalous sound data, which can be easily updated by using anomalous data. In this paper, we implement a new loss function based on metric learning to learn the distance relationship from each class centroid in feature space for the binary classification model. The proposed multi-task learning of the binary classification and the metric learning makes it possible to build the feature space where the within-class variance is minimized and the between-class variance is maximized while keeping normal and anomalous classes linearly separable. We also investigate the effectiveness of additionally using anomalous sound data for further improving the binary classification model. Our results showed that multi-task learning using binary classification and metric learning to consider the distance from each class centroid in the feature space is effective, and performance can be significantly improved by using even a small amount of anomalous data during training.

Wen-Chin Huang, Kazuhiro Kobayashi, Yu-Huai Peng et al.

We propose a new paradigm for maintaining speaker identity in dysarthric voice conversion (DVC). The poor quality of dysarthric speech can be greatly improved by statistical VC, but as the normal speech utterances of a dysarthria patient are nearly impossible to collect, previous work failed to recover the individuality of the patient. In light of this, we suggest a novel, two-stage approach for DVC, which is highly flexible in that no normal speech of the patient is required. First, a powerful parallel sequence-to-sequence model converts the input dysarthric speech into a normal speech of a reference speaker as an intermediate product, and a nonparallel, frame-wise VC model realized with a variational autoencoder then converts the speaker identity of the reference speech back to that of the patient while assumed to be capable of preserving the enhanced quality. We investigate several design options. Experimental evaluation results demonstrate the potential of our approach to improving the quality of the dysarthric speech while maintaining the speaker identity.

Patrick Lumban Tobing, Tomoki Toda

This paper presents a low-latency real-time (LLRT) non-parallel voice conversion (VC) framework based on cyclic variational autoencoder (CycleVAE) and multiband WaveRNN with data-driven linear prediction (MWDLP). CycleVAE is a robust non-parallel multispeaker spectral model, which utilizes a speaker-independent latent space and a speaker-dependent code to generate reconstructed/converted spectral features given the spectral features of an input speaker. On the other hand, MWDLP is an efficient and a high-quality neural vocoder that can handle multispeaker data and generate speech waveform for LLRT applications with CPU. To accommodate LLRT constraint with CPU, we propose a novel CycleVAE framework that utilizes mel-spectrogram as spectral features and is built with a sparse network architecture. Further, to improve the modeling performance, we also propose a novel fine-tuning procedure that refines the frame-rate CycleVAE network by utilizing the waveform loss from the MWDLP network. The experimental results demonstrate that the proposed framework achieves high-performance VC, while allowing for LLRT usage with a single-core of $2.1$--$2.7$ GHz CPU on a real-time factor of $0.87$--$0.95$, including input/output, feature extraction, on a frame shift of $10$ ms, a window length of $27.5$ ms, and $2$ lookup frames.

Patrick Lumban Tobing, Tomoki Toda

This paper presents a novel high-fidelity and low-latency universal neural vocoder framework based on multiband WaveRNN with data-driven linear prediction for discrete waveform modeling (MWDLP). MWDLP employs a coarse-fine bit WaveRNN architecture for 10-bit mu-law waveform modeling. A sparse gated recurrent unit with a relatively large size of hidden units is utilized, while the multiband modeling is deployed to achieve real-time low-latency usage. A novel technique for data-driven linear prediction (LP) with discrete waveform modeling is proposed, where the LP coefficients are estimated in a data-driven manner. Moreover, a novel loss function using short-time Fourier transform (STFT) for discrete waveform modeling with Gumbel approximation is also proposed. The experimental results demonstrate that the proposed MWDLP framework generates high-fidelity synthetic speech for seen and unseen speakers and/or language on 300 speakers training data including clean and noisy/reverberant conditions, where the number of training utterances is limited to 60 per speaker, while allowing for real-time low-latency processing using a single core of $\sim\!$ 2.1--2.7 GHz CPU with $\sim\!$ 0.57--0.64 real-time factor including input/output and feature extraction.

Tomoki Hayashi, Wen-Chin Huang, Kazuhiro Kobayashi et al.

This paper proposes a novel voice conversion (VC) method based on non-autoregressive sequence-to-sequence (NAR-S2S) models. Inspired by the great success of NAR-S2S models such as FastSpeech in text-to-speech (TTS), we extend the FastSpeech2 model for the VC problem. We introduce the convolution-augmented Transformer (Conformer) instead of the Transformer, making it possible to capture both local and global context information from the input sequence. Furthermore, we extend variance predictors to variance converters to explicitly convert the source speaker's prosody components such as pitch and energy into the target speaker. The experimental evaluation with the Japanese speaker dataset, which consists of male and female speakers of 1,000 utterances, demonstrates that the proposed model enables us to perform more stable, faster, and better conversion than autoregressive S2S (AR-S2S) models such as Tacotron2 and Transformer.

Reo Yoneyama, Yi-Chiao Wu, Tomoki Toda

We propose a unified approach to data-driven source-filter modeling using a single neural network for developing a neural vocoder capable of generating high-quality synthetic speech waveforms while retaining flexibility of the source-filter model to control their voice characteristics. Our proposed network called unified source-filter generative adversarial networks (uSFGAN) is developed by factorizing quasi-periodic parallel WaveGAN (QPPWG), one of the neural vocoders based on a single neural network, into a source excitation generation network and a vocal tract resonance filtering network by additionally implementing a regularization loss. Moreover, inspired by neural source filter (NSF), only a sinusoidal waveform is additionally used as the simplest clue to generate a periodic source excitation waveform while minimizing the effect of approximations in the source filter model. The experimental results demonstrate that uSFGAN outperforms conventional neural vocoders, such as QPPWG and NSF in both speech quality and pitch controllability.

Cheng-Hung Hu, Yi-Chiao Wu, Wen-Chin Huang et al.

This paper describes the AS-NU systems for two tracks in MultiSpeaker Multi-Style Voice Cloning Challenge (M2VoC). The first track focuses on using a small number of 100 target utterances for voice cloning, while the second track focuses on using only 5 target utterances for voice cloning. Due to the serious lack of data in the second track, we selected the speaker most similar to the target speaker from the training data of the TTS system, and used the speaker's utterances and the given 5 target utterances to fine-tune our model. The evaluation results show that our systems on the two tracks perform similarly in terms of quality, but there is still a clear gap between the similarity score of the second track and the similarity score of the first track.

Wen-Chin Huang, Chia-Hua Wu, Shang-Bao Luo et al.

We propose a simple method for automatic speech recognition (ASR) by fine-tuning BERT, which is a language model (LM) trained on large-scale unlabeled text data and can generate rich contextual representations. Our assumption is that given a history context sequence, a powerful LM can narrow the range of possible choices and the speech signal can be used as a simple clue. Hence, comparing to conventional ASR systems that train a powerful acoustic model (AM) from scratch, we believe that speech recognition is possible by simply fine-tuning a BERT model. As an initial study, we demonstrate the effectiveness of the proposed idea on the AISHELL dataset and show that stacking a very simple AM on top of BERT can yield reasonable performance.

Wen-Chin Huang, Yi-Chiao Wu, Tomoki Hayashi et al.

We present a novel approach to any-to-one (A2O) voice conversion (VC) in a sequence-to-sequence (seq2seq) framework. A2O VC aims to convert any speaker, including those unseen during training, to a fixed target speaker. We utilize vq-wav2vec (VQW2V), a discretized self-supervised speech representation that was learned from massive unlabeled data, which is assumed to be speaker-independent and well corresponds to underlying linguistic contents. Given a training dataset of the target speaker, we extract VQW2V and acoustic features to estimate a seq2seq mapping function from the former to the latter. With the help of a pretraining method and a newly designed postprocessing technique, our model can be generalized to only 5 min of data, even outperforming the same model trained with parallel data.