Yuxuan Wang

Qianqian Dong, Zhiying Huang, Qiao Tian et al.

We propose PolyVoice, a language model-based framework for speech-to-speech translation (S2ST) system. Our framework consists of two language models: a translation language model and a speech synthesis language model. We use discretized speech units, which are generated in a fully unsupervised way, and thus our framework can be used for unwritten languages. For the speech synthesis part, we adopt the existing VALL-E X approach and build a unit-based audio language model. This grants our framework the ability to preserve the voice characteristics and the speaking style of the original speech. We examine our system on Chinese $\rightarrow$ English and English $\rightarrow$ Spanish pairs. Experimental results show that our system can generate speech with high translation quality and audio quality. Speech samples are available at https://speechtranslation.github.io/polyvoice.

Giorgio Severi, Matthew Jagielski, Gökberk Yar et al.

Federated learning is a popular strategy for training models on distributed, sensitive data, while preserving data privacy. Prior work identified a range of security threats on federated learning protocols that poison the data or the model. However, federated learning is a networked system where the communication between clients and server plays a critical role for the learning task performance. We highlight how communication introduces another vulnerability surface in federated learning and study the impact of network-level adversaries on training federated learning models. We show that attackers dropping the network traffic from carefully selected clients can significantly decrease model accuracy on a target population. Moreover, we show that a coordinated poisoning campaign from a few clients can amplify the dropping attacks. Finally, we develop a server-side defense which mitigates the impact of our attacks by identifying and up-sampling clients likely to positively contribute towards target accuracy. We comprehensively evaluate our attacks and defenses on three datasets, assuming encrypted communication channels and attackers with partial visibility of the network.

Yukun Feng, Ming Tu, Rui Xia et al.

Recent studies have shown that using an external Language Model (LM) benefits the end-to-end Automatic Speech Recognition (ASR). However, predicting tokens that appear less frequently in the training set is still quite challenging. The long-tail prediction problems have been widely studied in many applications, but only been addressed by a few studies for ASR and LMs. In this paper, we propose a new memory augmented lookup dictionary based Transformer architecture for LM. The newly introduced lookup dictionary incorporates rich contextual information in training set, which is vital to correctly predict long-tail tokens. With intensive experiments on Chinese and English data sets, our proposed method is proved to outperform the baseline Transformer LM by a great margin on both word/character error rate and tail tokens error rate. This is achieved without impact on the decoding efficiency. Overall, we demonstrate the effectiveness of our proposed method in boosting the ASR decoding performance, especially for long-tail tokens.

Yicheng Feng, Yuxuan Wang, Jiazheng Liu et al.

Recently, various studies have leveraged Large Language Models (LLMs) to help decision-making and planning in environments, and try to align the LLMs' knowledge with the world conditions. Nonetheless, the capacity of LLMs to continuously acquire environmental knowledge and adapt in an open world remains uncertain. In this paper, we propose an approach to spur LLMs to explore the open world, gather experiences, and learn to improve their task-solving capabilities. In this approach, a multi-round feedback-revision mechanism is utilized to encourage LLMs to actively select appropriate revision actions guided by feedback information from the environment. This facilitates exploration and enhances the model's performance. Besides, we integrate sub-task relabeling to assist LLMs in maintaining consistency in sub-task planning and help the model learn the combinatorial nature between tasks, enabling it to complete a wider range of tasks through training based on the acquired exploration experiences. By evaluation in Minecraft, an open-ended sandbox world, we demonstrate that our approach LLaMA-Rider enhances the efficiency of the LLM in exploring the environment, and effectively improves the LLM's ability to accomplish more tasks through fine-tuning with merely 1.3k instances of collected data, showing minimal training costs compared to the baseline using reinforcement learning.

Yuxuan Wang, Hong Lyu

The information retrieval community has made significant progress in improving the efficiency of Dual Encoder (DE) dense passage retrieval systems, making them suitable for latency-sensitive settings. However, many proposed procedures are often too complex or resource-intensive, which makes it difficult for practitioners to adopt them or identify sources of empirical gains. Therefore, in this work, we propose a trivially simple recipe to serve as a baseline method for boosting the efficiency of DE retrievers leveraging an asymmetric architecture. Our results demonstrate that even a 2-layer, BERT-based query encoder can still retain 92.5% of the full DE performance on the BEIR benchmark via unsupervised distillation and proper student initialization. We hope that our findings will encourage the community to re-evaluate the trade-offs between method complexity and performance improvements.

Zhihao Du, Yuxuan Wang, Qian Chen et al.

In our previous work, we introduced CosyVoice, a multilingual speech synthesis model based on supervised discrete speech tokens. By employing progressive semantic decoding with two popular generative models, language models (LMs) and Flow Matching, CosyVoice demonstrated high prosody naturalness, content consistency, and speaker similarity in speech in-context learning. Recently, significant progress has been made in multi-modal large language models (LLMs), where the response latency and real-time factor of speech synthesis play a crucial role in the interactive experience. Therefore, in this report, we present an improved streaming speech synthesis model, CosyVoice 2, which incorporates comprehensive and systematic optimizations. Specifically, we introduce finite-scalar quantization to improve the codebook utilization of speech tokens. For the text-speech LM, we streamline the model architecture to allow direct use of a pre-trained LLM as the backbone. In addition, we develop a chunk-aware causal flow matching model to support various synthesis scenarios, enabling both streaming and non-streaming synthesis within a single model. By training on a large-scale multilingual dataset, CosyVoice 2 achieves human-parity naturalness, minimal response latency, and virtually lossless synthesis quality in the streaming mode. We invite readers to listen to the demos at https://funaudiollm.github.io/cosyvoice2.

Zhengpeng Shi, Hengli Li, Yanpeng Zhao et al.

AI models capable of comprehending humor hold real-world promise -- for example, enhancing engagement in human-machine interactions. To gauge and diagnose the capacity of multimodal large language models (MLLMs) for humor understanding, we introduce v-HUB, a novel visual-centric video humor understanding benchmark. v-HUB comprises a curated collection of minimally verbal short videos, sourced from classic silent films and online resources, and reflecting real-world scenarios where humor can be appreciated purely through visual cues. Each video clip is paired with rich annotations, including captions, descriptions, and explanations, supporting evaluation tasks like caption matching and humor explanation. To broaden its applicability, we further construct an open-ended video QA task, making it readily integrable into existing video understanding benchmarks. We evaluate a diverse set of MLLMs, from specialized Video-LLMs to versatile OmniLLMs that can process audio, covering both open-source and proprietary domains. The experimental results expose the difficulties MLLMs face in comprehending humor from visual cues alone. For example, all models exhibit a marked performance drop on caption matching when moving from text-based to video-based evaluation (without audio). Our findings also demonstrate that incorporating audio helps with video humor understanding, highlighting the informativeness of sound and the promise of integrating richer modalities for complex video understanding tasks.

Junyi Ao, Yuancheng Wang, Xiaohai Tian et al.

Speech encompasses a wealth of information, including but not limited to content, paralinguistic, and environmental information. This comprehensive nature of speech significantly impacts communication and is crucial for human-computer interaction. Chat-Oriented Large Language Models (LLMs), known for their general-purpose assistance capabilities, have evolved to handle multi-modal inputs, including speech. Although these models can be adept at recognizing and analyzing speech, they often fall short of generating appropriate responses. We argue that this is due to the lack of principles on task definition and model development, which requires open-source datasets and metrics suitable for model evaluation. To bridge the gap, we present SD-Eval, a benchmark dataset aimed at multidimensional evaluation of spoken dialogue understanding and generation. SD-Eval focuses on paralinguistic and environmental information and includes 7,303 utterances, amounting to 8.76 hours of speech data. The data is aggregated from eight public datasets, representing four perspectives: emotion, accent, age, and background sound. To assess the SD-Eval benchmark dataset, we implement three different models and construct a training set following a process similar to that of SD-Eval. The training set contains 1,052.72 hours of speech data and 724.4k utterances. We also conduct a comprehensive evaluation using objective evaluation methods (e.g. BLEU and ROUGE), subjective evaluations and LLM-based metrics for the generated responses. Models conditioned with paralinguistic and environmental information outperform their counterparts in both objective and subjective measures. Moreover, experiments demonstrate that LLM-based metrics show a higher correlation with human evaluation compared to traditional metrics. We open-source SD-Eval at https://github.com/amphionspace/SD-Eval.

Siyuan Feng, Ming Tu, Rui Xia et al.

Multilingual training is effective in improving low-resource ASR, which may partially be explained by phonetic representation sharing between languages. In end-to-end (E2E) ASR systems, graphemes are often used as basic modeling units, however graphemes may not be ideal for multilingual phonetic sharing. In this paper, we leverage International Phonetic Alphabet (IPA) based language-universal phonetic model to improve low-resource ASR performances, for the first time within the attention encoder-decoder architecture. We propose an adaptation method on the phonetic IPA model to further improve the proposed approach on extreme low-resource languages. Experiments carried out on the open-source MLS corpus and our internal databases show our approach outperforms baseline monolingual models and most state-of-the-art works. Our main approach and adaptation are effective on extremely low-resource languages, even within domain- and language-mismatched scenarios.

Haohe Liu, Qiuqiang Kong, Qiao Tian et al.

Speech restoration aims to remove distortions in speech signals. Prior methods mainly focus on single-task speech restoration (SSR), such as speech denoising or speech declipping. However, SSR systems only focus on one task and do not address the general speech restoration problem. In addition, previous SSR systems show limited performance in some speech restoration tasks such as speech super-resolution. To overcome those limitations, we propose a general speech restoration (GSR) task that attempts to remove multiple distortions simultaneously. Furthermore, we propose VoiceFixer, a generative framework to address the GSR task. VoiceFixer consists of an analysis stage and a synthesis stage to mimic the speech analysis and comprehension of the human auditory system. We employ a ResUNet to model the analysis stage and a neural vocoder to model the synthesis stage. We evaluate VoiceFixer with additive noise, room reverberation, low-resolution, and clipping distortions. Our baseline GSR model achieves a 0.499 higher mean opinion score (MOS) than the speech enhancement SSR model. VoiceFixer further surpasses the GSR baseline model on the MOS score by 0.256. Moreover, we observe that VoiceFixer generalizes well to severely degraded real speech recordings, indicating its potential in restoring old movies and historical speeches. The source code is available at https://github.com/haoheliu/voicefixer_main.

Zhihao Du, Changfeng Gao, Yuxuan Wang et al.

In our prior works, we introduced a scalable streaming speech synthesis model, CosyVoice 2, which integrates a large language model (LLM) and a chunk-aware flow matching (FM) model, and achieves low-latency bi-streaming speech synthesis and human-parity quality. Despite these advancements, CosyVoice 2 exhibits limitations in language coverage, domain diversity, data volume, text formats, and post-training techniques. In this paper, we present CosyVoice 3, an improved model designed for zero-shot multilingual speech synthesis in the wild, surpassing its predecessor in content consistency, speaker similarity, and prosody naturalness. Key features of CosyVoice 3 include: 1) A novel speech tokenizer to improve prosody naturalness, developed via supervised multi-task training, including automatic speech recognition, speech emotion recognition, language identification, audio event detection, and speaker analysis. 2) A new differentiable reward model for post-training applicable not only to CosyVoice 3 but also to other LLM-based speech synthesis models. 3) Dataset Size Scaling: Training data is expanded from ten thousand hours to one million hours, encompassing 9 languages and 18 Chinese dialects across various domains and text formats. 4) Model Size Scaling: Model parameters are increased from 0.5 billion to 1.5 billion, resulting in enhanced performance on our multilingual benchmark due to the larger model capacity. These advancements contribute significantly to the progress of speech synthesis in the wild. We encourage readers to listen to the demo at https://funaudiollm.github.io/cosyvoice3.

Ziyang Ma, Yinghao Ma, Yanqiao Zhu et al.

We introduce MMAR, a new benchmark designed to evaluate the deep reasoning capabilities of Audio-Language Models (ALMs) across massive multi-disciplinary tasks. MMAR comprises 1,000 meticulously curated audio-question-answer triplets, collected from real-world internet videos and refined through iterative error corrections and quality checks to ensure high quality. Unlike existing benchmarks that are limited to specific domains of sound, music, or speech, MMAR extends them to a broad spectrum of real-world audio scenarios, including mixed-modality combinations of sound, music, and speech. Each question in MMAR is hierarchically categorized across four reasoning layers: Signal, Perception, Semantic, and Cultural, with additional sub-categories within each layer to reflect task diversity and complexity. To further foster research in this area, we annotate every question with a Chain-of-Thought (CoT) rationale to promote future advancements in audio reasoning. Each item in the benchmark demands multi-step deep reasoning beyond surface-level understanding. Moreover, a part of the questions requires graduate-level perceptual and domain-specific knowledge, elevating the benchmark's difficulty and depth. We evaluate MMAR using a broad set of models, including Large Audio-Language Models (LALMs), Large Audio Reasoning Models (LARMs), Omni Language Models (OLMs), Large Language Models (LLMs), and Large Reasoning Models (LRMs), with audio caption inputs. The performance of these models on MMAR highlights the benchmark's challenging nature, and our analysis further reveals critical limitations of understanding and reasoning capabilities among current models. We hope MMAR will serve as a catalyst for future advances in this important but little-explored area.

Philip Anastassiou, Zhenyu Tang, Kainan Peng et al. · bytedance

We present VoiceShop, a novel speech-to-speech framework that can modify multiple attributes of speech, such as age, gender, accent, and speech style, in a single forward pass while preserving the input speaker's timbre. Previous works have been constrained to specialized models that can only edit these attributes individually and suffer from the following pitfalls: the magnitude of the conversion effect is weak, there is no zero-shot capability for out-of-distribution speakers, or the synthesized outputs exhibit undesirable timbre leakage. Our work proposes solutions for each of these issues in a simple modular framework based on a conditional diffusion backbone model with optional normalizing flow-based and sequence-to-sequence speaker attribute-editing modules, whose components can be combined or removed during inference to meet a wide array of tasks without additional model finetuning. Audio samples are available at \url{https://voiceshopai.github.io}.

Yuxuan Wang, Xuanyu Yi, Haohan Weng et al.

Triangle meshes are fundamental to 3D applications, enabling efficient modification and rasterization while maintaining compatibility with standard rendering pipelines. However, current automatic mesh generation methods typically rely on intermediate representations that lack the continuous surface quality inherent to meshes. Converting these representations into meshes produces dense, suboptimal outputs. Although recent autoregressive approaches demonstrate promise in directly modeling mesh vertices and faces, they are constrained by the limitation in face count, scalability, and structural fidelity. To address these challenges, we propose Nautilus, a locality-aware autoencoder for artist-like mesh generation that leverages the local properties of manifold meshes to achieve structural fidelity and efficient representation. Our approach introduces a novel tokenization algorithm that preserves face proximity relationships and compresses sequence length through locally shared vertices and edges, enabling the generation of meshes with an unprecedented scale of up to 5,000 faces. Furthermore, we develop a Dual-stream Point Conditioner that provides multi-scale geometric guidance, ensuring global consistency and local structural fidelity by capturing fine-grained geometric features. Extensive experiments demonstrate that Nautilus significantly outperforms state-of-the-art methods in both fidelity and scalability. The project page is at https://nautilusmeshgen.github.io.

Biwen Lei, Yang Li, Xinhai Liu et al.

The creation of high-quality 3D assets, a cornerstone of modern game development, has long been characterized by labor-intensive and specialized workflows. This paper presents Hunyuan3D Studio, an end-to-end AI-powered content creation platform designed to revolutionize the game production pipeline by automating and streamlining the generation of game-ready 3D assets. At its core, Hunyuan3D Studio integrates a suite of advanced neural modules (such as Part-level 3D Generation, Polygon Generation, Semantic UV, etc.) into a cohesive and user-friendly system. This unified framework allows for the rapid transformation of a single concept image or textual description into a fully-realized, production-quality 3D model complete with optimized geometry and high-fidelity PBR textures. We demonstrate that assets generated by Hunyuan3D Studio are not only visually compelling but also adhere to the stringent technical requirements of contemporary game engines, significantly reducing iteration time and lowering the barrier to entry for 3D content creation. By providing a seamless bridge from creative intent to technical asset, Hunyuan3D Studio represents a significant leap forward for AI-assisted workflows in game development and interactive media.

Daniel Chin, Yuxuan Wang, Gus Xia

Large Language Model (LLM) -in-the-loop applications have been shown to effectively interpret the human user's commands, make plans, and operate external tools/systems accordingly. Still, the operation scope of the LLM agent is limited to passively following the user, requiring the user to frame his/her needs with regard to the underlying tools/systems. We note that the potential of an LLM-Agent User Interface (LAUI) is much greater. A user mostly ignorant to the underlying tools/systems should be able to work with a LAUI to discover an emergent workflow. Contrary to the conventional way of designing an explorable GUI to teach the user a predefined set of ways to use the system, in the ideal LAUI, the LLM agent is initialized to be proficient with the system, proactively studies the user and his/her needs, and proposes new interaction schemes to the user. To illustrate LAUI, we present Flute X GPT, a concrete example using an LLM agent, a prompt manager, and a flute-tutoring multi-modal software-hardware system to facilitate the complex, real-time user experience of learning to play the flute.

Yijie Zheng, Bangjun Xiao, Lei Shi et al.

Multimodal large language models (MLLMs), such as GPT-4o, are garnering significant attention. During the exploration of MLLM training, we identified Modality Composition Incoherence, a phenomenon that the proportion of a certain modality varies dramatically across different examples. It exacerbates the challenges of addressing mini-batch imbalances, which lead to uneven GPU utilization between Data Parallel (DP) instances and severely degrades the efficiency and scalability of MLLM training, ultimately affecting training speed and hindering further research on MLLMs. To address these challenges, we introduce OrchMLLM, a comprehensive framework designed to mitigate the inefficiencies in MLLM training caused by Modality Composition Incoherence. First, we propose Batch Post-Balancing Dispatcher, a technique that efficiently eliminates mini-batch imbalances in sequential data. Additionally, we integrate MLLM Global Orchestrator into the training framework to orchestrate multimodal data and tackle the issues arising from Modality Composition Incoherence. We evaluate OrchMLLM across various MLLM sizes, demonstrating its efficiency and scalability. Experimental results reveal that OrchMLLM achieves a Model FLOPs Utilization (MFU) of $41.6\%$ when training an 84B MLLM with three modalities on $2560$ H100 GPUs, outperforming Megatron-LM by up to $3.1\times$ in throughput.

Junyi Ao, Dekun Chen, Xiaohai Tian et al.

Large Language Models (LLMs) have recently shown remarkable ability to process not only text but also multimodal inputs such as speech and audio. However, most existing models primarily focus on analyzing input signals using text instructions, overlooking scenarios in which speech instructions and audio are mixed and serve as inputs to the model. To address these challenges, we introduce Solla, a novel framework designed to understand speech-based questions and hear the acoustic context concurrently. Solla incorporates an audio tagging module to effectively identify and represent audio events, as well as an ASR-assisted prediction method to improve comprehension of spoken content. To rigorously evaluate Solla and other publicly available models, we propose a new benchmark dataset called SA-Eval, which includes three tasks: audio event classification, audio captioning, and audio question answering. SA-Eval has diverse speech instruction with various speaking styles, encompassing two difficulty levels, easy and hard, to capture the range of real-world acoustic conditions. Experimental results show that Solla performs on par with or outperforms baseline models on both the easy and hard test sets, underscoring its effectiveness in jointly understanding speech and audio.

Junjie Yang, Yuhao Yan, Gang Wu et al.

As Vision-Language Models (VLMs) increasingly gain traction in medical applications, clinicians are progressively expecting AI systems not only to generate textual diagnoses but also to produce corresponding medical images that integrate seamlessly into authentic clinical workflows. Despite the growing interest, existing medical visual benchmarks present notable limitations. They often rely on ambiguous queries that lack sufficient relevance to image content, oversimplify complex diagnostic reasoning into closed-ended shortcuts, and adopt a text-centric evaluation paradigm that overlooks the importance of image generation capabilities. To address these challenges, we introduce MedGEN-Bench, a comprehensive multimodal benchmark designed to advance medical AI research. MedGEN-Bench comprises 6,422 expert-validated image-text pairs spanning six imaging modalities, 16 clinical tasks, and 28 subtasks. It is structured into three distinct formats: Visual Question Answering, Image Editing, and Contextual Multimodal Generation. What sets MedGEN-Bench apart is its focus on contextually intertwined instructions that necessitate sophisticated cross-modal reasoning and open-ended generative outputs, moving beyond the constraints of multiple-choice formats. To evaluate the performance of existing systems, we employ a novel three-tier assessment framework that integrates pixel-level metrics, semantic text analysis, and expert-guided clinical relevance scoring. Using this framework, we systematically assess 10 compositional frameworks, 3 unified models, and 5 VLMs.

Van Tung Pham, Yist Lin, Tao Han et al.

Recent works have shown promising results in connecting speech encoders to large language models (LLMs) for speech recognition. However, several limitations persist, including limited fine-tuning options, a lack of mechanisms to enforce speech-text alignment, and high insertion errors especially in domain mismatch conditions. This paper presents a comprehensive solution to address these issues. We begin by investigating more thoughtful fine-tuning schemes. Next, we propose a matching loss to enhance alignment between modalities. Finally, we explore training and inference methods to mitigate high insertion errors. Experimental results on the Librispeech corpus demonstrate that partially fine-tuning the encoder and LLM using parameter-efficient methods, such as LoRA, is the most cost-effective approach. Additionally, the matching loss improves modality alignment, enhancing performance. The proposed training and inference methods significantly reduce insertion errors.

Yuxuan Wang, Jianghui Wang, Dongyan Zhao et al.

We introduce CDBERT, a new learning paradigm that enhances the semantics understanding ability of the Chinese PLMs with dictionary knowledge and structure of Chinese characters. We name the two core modules of CDBERT as Shuowen and Jiezi, where Shuowen refers to the process of retrieving the most appropriate meaning from Chinese dictionaries and Jiezi refers to the process of enhancing characters' glyph representations with structure understanding. To facilitate dictionary understanding, we propose three pre-training tasks, i.e., Masked Entry Modeling, Contrastive Learning for Synonym and Antonym, and Example Learning. We evaluate our method on both modern Chinese understanding benchmark CLUE and ancient Chinese benchmark CCLUE. Moreover, we propose a new polysemy discrimination task PolyMRC based on the collected dictionary of ancient Chinese. Our paradigm demonstrates consistent improvements on previous Chinese PLMs across all tasks. Moreover, our approach yields significant boosting on few-shot setting of ancient Chinese understanding.

Siyuan Feng, Ming Tu, Rui Xia et al.

We improve low-resource ASR by integrating the ideas of multilingual training and self-supervised learning. Concretely, we leverage an International Phonetic Alphabet (IPA) multilingual model to create frame-level pseudo labels for unlabeled speech, and use these pseudo labels to guide hidden-unit BERT (HuBERT) based speech pretraining in a phonetically-informed manner. The experiments on the Multilingual Speech (MLS) Corpus show that the proposed approach consistently outperforms the standard HuBERT on all the target languages. Moreover, on 3 of the 4 languages, comparing to the standard HuBERT, the approach performs better, meanwhile is able to save supervised training data by 1.5k hours (75%) at most. Our approach outperforms most of the state of the arts, with much less pretraining data in terms of hours and language diversity. Compared to XLSR-53 and a retraining based multilingual method, our approach performs better with full and limited finetuning data scenarios.

Zelin Ying, Chen Li, Yu Dong et al.

The front-end is a critical component of English text-to-speech (TTS) systems, responsible for extracting linguistic features that are essential for a text-to-speech model to synthesize speech, such as prosodies and phonemes. The English TTS front-end typically consists of a text normalization (TN) module, a prosody word prosody phrase (PWPP) module, and a grapheme-to-phoneme (G2P) module. However, current research on the English TTS front-end focuses solely on individual modules, neglecting the interdependence between them and resulting in sub-optimal performance for each module. Therefore, this paper proposes a unified front-end framework that captures the dependencies among the English TTS front-end modules. Extensive experiments have demonstrated that the proposed method achieves state-of-the-art (SOTA) performance in all modules.

Maokui He, Xiang Lv, Weilin Zhou et al.

We propose two improvements to target-speaker voice activity detection (TS-VAD), the core component in our proposed speaker diarization system that was submitted to the 2022 Multi-Channel Multi-Party Meeting Transcription (M2MeT) challenge. These techniques are designed to handle multi-speaker conversations in real-world meeting scenarios with high speaker-overlap ratios and under heavy reverberant and noisy condition. First, for data preparation and augmentation in training TS-VAD models, speech data containing both real meetings and simulated indoor conversations are used. Second, in refining results obtained after TS-VAD based decoding, we perform a series of post-processing steps to improve the VAD results needed to reduce diarization error rates (DERs). Tested on the ALIMEETING corpus, the newly released Mandarin meeting dataset used in M2MeT, we demonstrate that our proposed system can decrease the DER by up to 66.55/60.59% relatively when compared with classical clustering based diarization on the Eval/Test set.

Chenxu Hu, Qiao Tian, Tingle Li et al.

Dubbing is a post-production process of re-recording actors' dialogues, which is extensively used in filmmaking and video production. It is usually performed manually by professional voice actors who read lines with proper prosody, and in synchronization with the pre-recorded videos. In this work, we propose Neural Dubber, the first neural network model to solve a novel automatic video dubbing (AVD) task: synthesizing human speech synchronized with the given video from the text. Neural Dubber is a multi-modal text-to-speech (TTS) model that utilizes the lip movement in the video to control the prosody of the generated speech. Furthermore, an image-based speaker embedding (ISE) module is developed for the multi-speaker setting, which enables Neural Dubber to generate speech with a reasonable timbre according to the speaker's face. Experiments on the chemistry lecture single-speaker dataset and LRS2 multi-speaker dataset show that Neural Dubber can generate speech audios on par with state-of-the-art TTS models in terms of speech quality. Most importantly, both qualitative and quantitative evaluations show that Neural Dubber can control the prosody of synthesized speech by the video, and generate high-fidelity speech temporally synchronized with the video. Our project page is at https://tsinghua-mars-lab.github.io/NeuralDubber/ .

Dongyang Dai, Yuanzhe Chen, Li Chen et al.

With the increasing popularity of speech synthesis products, the industry has put forward more requirements for personalized speech synthesis: (1) How to use low-resource, easily accessible data to clone a person's voice. (2) How to clone a person's voice while controlling the style and prosody. To solve the above two problems, we proposed the Hieratron model framework in which the prosody and timbre are modeled separately using two modules, therefore, the independent control of timbre and the other characteristics of audio can be achieved while generating speech. The practice shows that, for very limited target speaker data in the wild, Hieratron has obvious advantages over the traditional method, in addition to controlling the style and language of the generated speech, the mean opinion score on speech quality of the generated speech has also been improved by more than 0.2 points.

Yuxuan Wang, Maokui He, Shutong Niu et al.

This system description describes our submission system to the Third DIHARD Speech Diarization Challenge. Besides the traditional clustering based system, the innovation of our system lies in the combination of various front-end techniques to solve the diarization problem, including speech separation and target-speaker based voice activity detection (TS-VAD), combined with iterative data purification. We also adopted audio domain classification to design domain-dependent processing. Finally, we performed post processing to do system fusion and selection. Our best system achieved DERs of 11.30% in track 1 and 16.78% in track 2 on evaluation set, respectively.

Qiuqiang Kong, Haohe Liu, Xingjian Du et al.

Speech enhancement is a task to improve the intelligibility and perceptual quality of degraded speech signal. Recently, neural networks based methods have been applied to speech enhancement. However, many neural network based methods require noisy and clean speech pairs for training. We propose a speech enhancement framework that can be trained with large-scale weakly labelled AudioSet dataset. Weakly labelled data only contain audio tags of audio clips, but not the onset or offset times of speech. We first apply pretrained audio neural networks (PANNs) to detect anchor segments that contain speech or sound events in audio clips. Then, we randomly mix two detected anchor segments containing speech and sound events as a mixture, and build a conditional source separation network using PANNs predictions as soft conditions for speech enhancement. In inference, we input a noisy speech signal with the one-hot encoding of "Speech" as a condition to the trained system to predict enhanced speech. Our system achieves a PESQ of 2.28 and an SSNR of 8.75 dB on the VoiceBank-DEMAND dataset, outperforming the previous SEGAN system of 2.16 and 7.73 dB respectively.

Wenjie Li, Benlai Tang, Xiang Yin et al.

Accent conversion (AC) transforms a non-native speaker's accent into a native accent while maintaining the speaker's voice timbre. In this paper, we propose approaches to improving accent conversion applicability, as well as quality. First of all, we assume no reference speech is available at the conversion stage, and hence we employ an end-to-end text-to-speech system that is trained on native speech to generate native reference speech. To improve the quality and accent of the converted speech, we introduce reference encoders which make us capable of utilizing multi-source information. This is motivated by acoustic features extracted from native reference and linguistic information, which are complementary to conventional phonetic posteriorgrams (PPGs), so they can be concatenated as features to improve a baseline system based only on PPGs. Moreover, we optimize model architecture using GMM-based attention instead of windowed attention to elevate synthesized performance. Experimental results indicate when the proposed techniques are applied the integrated system significantly raises the scores of acoustic quality (30$\%$ relative increase in mean opinion score) and native accent (68$\%$ relative preference) while retaining the voice identity of the non-native speaker.

Junjie Pan, Xiang Yin, Zhiling Zhang et al.

In Mandarin text-to-speech (TTS) system, the front-end text processing module significantly influences the intelligibility and naturalness of synthesized speech. Building a typical pipeline-based front-end which consists of multiple individual components requires extensive efforts. In this paper, we proposed a unified sequence-to-sequence front-end model for Mandarin TTS that converts raw texts to linguistic features directly. Compared to the pipeline-based front-end, our unified front-end can achieve comparable performance in polyphone disambiguation and prosody word prediction, and improve intonation phrase prediction by 0.0738 in F1 score. We also implemented the unified front-end with Tacotron and WaveRNN to build a Mandarin TTS system. The synthesized speech by that got a comparable MOS (4.38) with the pipeline-based front-end (4.37) and close to human recordings (4.49).

Wei-Ning Hsu, Yu Zhang, Ron J. Weiss et al.

This paper proposes a neural sequence-to-sequence text-to-speech (TTS) model which can control latent attributes in the generated speech that are rarely annotated in the training data, such as speaking style, accent, background noise, and recording conditions. The model is formulated as a conditional generative model based on the variational autoencoder (VAE) framework, with two levels of hierarchical latent variables. The first level is a categorical variable, which represents attribute groups (e.g. clean/noisy) and provides interpretability. The second level, conditioned on the first, is a multivariate Gaussian variable, which characterizes specific attribute configurations (e.g. noise level, speaking rate) and enables disentangled fine-grained control over these attributes. This amounts to using a Gaussian mixture model (GMM) for the latent distribution. Extensive evaluation demonstrates its ability to control the aforementioned attributes. In particular, we train a high-quality controllable TTS model on real found data, which is capable of inferring speaker and style attributes from a noisy utterance and use it to synthesize clean speech with controllable speaking style.

Yu-An Chung, Yuxuan Wang, Wei-Ning Hsu et al.

Although end-to-end text-to-speech (TTS) models such as Tacotron have shown excellent results, they typically require a sizable set of high-quality <text, audio> pairs for training, which are expensive to collect. In this paper, we propose a semi-supervised training framework to improve the data efficiency of Tacotron. The idea is to allow Tacotron to utilize textual and acoustic knowledge contained in large, publicly-available text and speech corpora. Importantly, these external data are unpaired and potentially noisy. Specifically, first we embed each word in the input text into word vectors and condition the Tacotron encoder on them. We then use an unpaired speech corpus to pre-train the Tacotron decoder in the acoustic domain. Finally, we fine-tune the model using available paired data. We demonstrate that the proposed framework enables Tacotron to generate intelligible speech using less than half an hour of paired training data.

Daisy Stanton, Yuxuan Wang, RJ Skerry-Ryan

Global Style Tokens (GSTs) are a recently-proposed method to learn latent disentangled representations of high-dimensional data. GSTs can be used within Tacotron, a state-of-the-art end-to-end text-to-speech synthesis system, to uncover expressive factors of variation in speaking style. In this work, we introduce the Text-Predicted Global Style Token (TP-GST) architecture, which treats GST combination weights or style embeddings as "virtual" speaking style labels within Tacotron. TP-GST learns to predict stylistic renderings from text alone, requiring neither explicit labels during training nor auxiliary inputs for inference. We show that, when trained on a dataset of expressive speech, our system generates audio with more pitch and energy variation than two state-of-the-art baseline models. We further demonstrate that TP-GSTs can synthesize speech with background noise removed, and corroborate these analyses with positive results on human-rated listener preference audiobook tasks. Finally, we demonstrate that multi-speaker TP-GST models successfully factorize speaker identity and speaking style. We provide a website with audio samples for each of our findings.

Wanxiang Che, Yijia Liu, Yuxuan Wang et al.

This paper describes our system (HIT-SCIR) submitted to the CoNLL 2018 shared task on Multilingual Parsing from Raw Text to Universal Dependencies. We base our submission on Stanford's winning system for the CoNLL 2017 shared task and make two effective extensions: 1) incorporating deep contextualized word embeddings into both the part of speech tagger and parser; 2) ensembling parsers trained with different initialization. We also explore different ways of concatenating treebanks for further improvements. Experimental results on the development data show the effectiveness of our methods. In the final evaluation, our system was ranked first according to LAS (75.84%) and outperformed the other systems by a large margin.

RJ Skerry-Ryan, Eric Battenberg, Ying Xiao et al.

We present an extension to the Tacotron speech synthesis architecture that learns a latent embedding space of prosody, derived from a reference acoustic representation containing the desired prosody. We show that conditioning Tacotron on this learned embedding space results in synthesized audio that matches the prosody of the reference signal with fine time detail even when the reference and synthesis speakers are different. Additionally, we show that a reference prosody embedding can be used to synthesize text that is different from that of the reference utterance. We define several quantitative and subjective metrics for evaluating prosody transfer, and report results with accompanying audio samples from single-speaker and 44-speaker Tacotron models on a prosody transfer task.

Yuxuan Wang, Daisy Stanton, Yu Zhang et al.

In this work, we propose "global style tokens" (GSTs), a bank of embeddings that are jointly trained within Tacotron, a state-of-the-art end-to-end speech synthesis system. The embeddings are trained with no explicit labels, yet learn to model a large range of acoustic expressiveness. GSTs lead to a rich set of significant results. The soft interpretable "labels" they generate can be used to control synthesis in novel ways, such as varying speed and speaking style - independently of the text content. They can also be used for style transfer, replicating the speaking style of a single audio clip across an entire long-form text corpus. When trained on noisy, unlabeled found data, GSTs learn to factorize noise and speaker identity, providing a path towards highly scalable but robust speech synthesis.

Yuxuan Wang, RJ Skerry-Ryan, Ying Xiao et al.

Prosodic modeling is a core problem in speech synthesis. The key challenge is producing desirable prosody from textual input containing only phonetic information. In this preliminary study, we introduce the concept of "style tokens" in Tacotron, a recently proposed end-to-end neural speech synthesis model. Using style tokens, we aim to extract independent prosodic styles from training data. We show that without annotation data or an explicit supervision signal, our approach can automatically learn a variety of prosodic variations in a purely data-driven way. Importantly, each style token corresponds to a fixed style factor regardless of the given text sequence. As a result, we can control the prosodic style of synthetic speech in a somewhat predictable and globally consistent way.

Yuxuan Wang, RJ Skerry-Ryan, Daisy Stanton et al.

A text-to-speech synthesis system typically consists of multiple stages, such as a text analysis frontend, an acoustic model and an audio synthesis module. Building these components often requires extensive domain expertise and may contain brittle design choices. In this paper, we present Tacotron, an end-to-end generative text-to-speech model that synthesizes speech directly from characters. Given <text, audio> pairs, the model can be trained completely from scratch with random initialization. We present several key techniques to make the sequence-to-sequence framework perform well for this challenging task. Tacotron achieves a 3.82 subjective 5-scale mean opinion score on US English, outperforming a production parametric system in terms of naturalness. In addition, since Tacotron generates speech at the frame level, it's substantially faster than sample-level autoregressive methods.