A multilingual zero-shot TTS system that supports 14 languages and performs both intra-lingual and cross-lingual reference cloning without requiring transcripts of audio prompts, which achieves the best average overall rank in human evaluation among recent open-source and commercial systems.
Abstract
Recent advances in zero-shot text-to-speech (TTS) have substantially improved speech quality and voice cloning fidelity. However, many zero-shot TTS systems still depend on audio prompt transcripts at inference time. This dependency limits cross-lingual voice cloning, since in-the-wild reference audio is often untranscribed. In this technical report, we present Confucius4-TTS, a multilingual zero-shot TTS system that supports 14 languages and performs both intra-lingual and cross-lingual reference cloning without requiring transcripts of audio prompts. Confucius4-TTS follows a two-stage architecture, consisting of text-to-semantic (T2S) and semantic-to-acoustic (S2A) modules. The LLM-based T2S module uses a learnable speaker encoder to extract timbre features from self-supervised speech representations, and the conditional flow-matching S2A module converts the predicted semantic tokens into mel-spectrograms. The same model also supports continuation cloning when a reference transcript is available. Confucius4-TTS is trained on large-scale multilingual speech data. It achieves high intelligibility and speaker similarity on public benchmarks. On the CV3-Eval cross-lingual benchmark, Confucius4-TTS obtains an average WER of 3.73% across six directions. On our internal cross-lingual set, it achieves the best average overall rank in human evaluation among recent open-source and commercial systems. We release code, model checkpoints, and demos at https://github.com/netease-youdao/Confucius4-TTS.
Semantic speech tokens should preserve linguistic content while suppressing speaker- and duration-dependent variation inherited from acoustic inputs. We propose Iterative Semantic Token Purification (ISTP), an alternating speech-to-unit (S2U) and text-to-unit (T2U) training procedure guided by text predictability. Starting from an initial S2U tokenizer, each iteration trains a T2U model on its deduplicated token sequences. The decoded T2U predictions then serve as connectionist temporal classification targets for a newly initialized S2U model, whose outputs supervise the next T2U model. This cycle progressively aligns the two token generators and biases the token space toward information recoverable from text. Experiments on Mandarin and English show substantially improved S2U--T2U agreement. Independently trained de-tokenizers further show that the refined S2U and T2U tokens retain sufficient content for high-intelligibility voice conversion and text-to-speech synthesis. In voice conversion, the generated speaking rate follows the reference more closely. The refined tokens also exhibit substantially improved cross-speaker consistency and reduced probe-recoverable speaker information.
Hanlin Zhang, Daxin Tan, Dehua Tao et al.· 0 citations
Experimental results show that VALL-E outperforms the state-of-the-art zero-shot TTS system in terms of speech naturalness and speaker similarity and could preserve the speaker’s emotion and acoustic environment from the prompt in synthesis.
Detecting hate speech in low-resource and unseen languages remains challenging due to limited labeled data and linguistic diversity. This paper presents a comparative study of zero-shot cross-lingual transfer for hate speech detection using two multilingual transformer models: mDeBERTa-v3 and XLM-RoBERTa. To the best of our knowledge, mDeBERTa-v3 has not been previously used by researchers for zero-shot cross-lingual hate speech detection, making this the first study to evaluate its capabilities in this task. Furthermore, we introduce new unseen languages that have not been studied before in this context, including Hebrew, Amharic, and Swahili, alongside other languages such as Indonesian, Danish Portuguese, Turkish, French, and Russian. We evaluate model performance under three training scenarios: a single source language (Turkish), semantically similar language clusters, and multiple clusters from different language families. Experimental results show that mDeBERTa-v3 consistently outperforms XLM-RoBERTa in zero-shot settings. The most notable improvement is observed for Hebrew, where the macro F1 score increases from 0.39 (XLM-RoBERTa) to 0.71 (mDeBERTa-v3), a gain of 0.32. Substantial gains are also seen for Amharic (0.52 → 0.73, +0.21), Indonesian (0.57 → 0.71, +0.14), and Swahili (0.65 → 0.75, +0.10). Across all experimental conditions, mDeBERTa-v3 achieves average macro F1 gains ranging from 0.04 to 0.19, with statistical significance (p < 0.02). The model’s advantage is attributed to its disentangled attention mechanism, which enables better generalization across typologically distant languages. These findings establish mDeBERTa-v3 as a novel and more robust architecture for zero-shot cross-lingual hate speech detection, particularly for previously unexplored low-resource languages.
Ghadeer Al-Badani, M. Alsurori, Akram Alsubari· 2026 6th International Confe...· 0 citations
Empirical analyses show that MoSE improves high-, medium-, and low-resource languages simultaneously, with the largest gains on low-resource speech, thereby breaking the curse of multilinguality without compromising high-resource performance.
Yexing Du, Kaiyuan Liu, Youcheng Pan et al.· 0 citations
Discrete speech tokenizers aim to disentangle semantic from acoustic information, yet targets from self-supervised learning (SSL) models like HuBERT retain non-linguistic variation: speaker identity, prosody, and channel conditions leak into the tokens, inflating entropy. Our key insight is that when enough speakers utter the same words under varying conditions, linguistic content is the only shared factor. We propose PINT (Parallel INvariant Tokenization), which fine-tunes an SSL encoder with alignment losses across parallel utterances and augmentations to distill this shared residual. PINT collapses identical words onto consistent token sequences, drastically reducing conditional entropy. Unlike ASR text, PINT tokens preserve frame-level temporal grounding and serve as drop-in semantic targets for audio codecs. Experiments show a 98.7% relative reduction in speaker probe accuracy (93.1% to 1.2%), a 42% lower ABX error rate, and 27-30% lower LM perplexity versus baselines, confirming that the right invariance is key to efficient learning.
Laurin Wagner, Bernhard Thallinger, Miroslav Stankovič et al.· arXiv.org· 2 citations
In current zero-shot text-to-speech systems, conventional semantic tokenizers are typically optimized using supervised automatic speech recognition or self-supervised learning objectives. However, due to the inherent nature of speech, semantic and acoustic information cannot be completely decoupled, and ASR-based tokenizers discard acoustic details to focus on linguistic content; models relying on them usually struggle to achieve optimal speaker similarity. Furthermore, these tokenizers are optimized independently and lack direct supervision from downstream acoustic generation tasks. This isolated training creates a feature gap between the extracted discrete tokens and the continuous space required by acoustic models, fundamentally bottlenecking the upper bound of synthesis quality. To bridge this gap, we propose Phoenix TTS, a unified framework that tightly couples representation learning with generative acoustic modeling. Specifically, our speech tokenizer is optimized to reconstruct self-supervised features to maintain semantic richness, while simultaneously receiving direct supervision from a Flow Matching training loss. Through this joint training paradigm, the extracted discrete tokens successfully preserve essential semantic information and natively align with the feature space of the downstream Flow Matching model. Comprehensive evaluations highlight the efficiency and effectiveness of Phoenix TTS. Trained on 110K hours of data, the system achieves excellent speech intelligibility, yielding WER that consistently falls below that of ground-truth recordings. Simultaneously, it maintains robust zero-shot speaker similarity, rivaling or outperforming several prominent large-scale baselines. Furthermore, as an advantageous byproduct of this unified training, the learned tokenizer can be seamlessly adapted to zero-shot voice conversion tasks without requiring task-specific fine-tuning.
Peijie Chen, Zhuanling Zha, Zhipeng Nie et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
