arXiv subjects

Runchuan Ye

Publications and source records attributed to Runchuan Ye.

7 recordsLinked to original sources

arXiv2026

VoxCPM2 Technical Report

We present VoxCPM2, a https://info.arxiv.org/help/prep#abstractsfully open-source multilingual and controllable speech generation foundation model that extends the hierarchical diffusion-autoregressive modeling paradigm of VoxCPM. VoxCPM2 advances the framework in three key dimensions: (i) capability, by unifying 30 languages, 9 Chinese dialects, natural-language voice design, style-controllable voice cloning, and high-fidelity continuation cloning within a single backbone; (ii) quality, through an asymmetric AudioVAE that encodes at 16 kHz and reconstructs at 48 kHz, enabling implicit super-resolution with high encoding efficiency; and (iii) scale, by jointly scaling the model to 2B parameters and the training data to over 2 million hours of multilingual speech. To support these diverse capabilities within one model, we introduce a unified sequence organization that expresses all generation modes through different arrangements of the same input building blocks, allowing joint training under a single set of parameters and objective. VoxCPM2 achieves state-of-the-art or competitive performance on public zero-shot and instruction-following TTS benchmarks. On our internal 30-language evaluation set, it attains an average WER of 1.68%. These results demonstrate that hierarchical continuous-latent modeling, without relying on any external discrete speech tokenizer, offers a viable and powerful foundation for large-scale multilingual and controllable speech generation. The model weights, fine-tuning code, and inference tools are publicly released under the Apache 2.0 license to foster community research and development.

cs.SD↗

arXiv2025

VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning

Generative models for speech synthesis face a fundamental trade-off: discrete tokens ensure stability but sacrifice expressivity, while continuous signals retain acoustic richness but suffer from error accumulation due to task entanglement. This challenge has driven the field towards multi-stage pipelines that rely on pre-trained speech tokenizers, but these create a semantic-acoustic divide, limiting holistic and expressive speech generation. We resolve these dilemma through hierarchical semantic-acoustic modeling with semi-discrete residual representations and present a novel tokenizer-free TTS model VoxCPM. Our framework introduces a differentiable quantization bottleneck that induces natural specialization: a Text-Semantic Language Model (TSLM) generates semantic-prosodic plans, while a Residual Acoustic Model (RALM) recovers fine-grained acoustic details. This hierarchical semantic-acoustic representation guides a local diffusion-based decoder to generate high-fidelity speech latents. Critically, the entire architecture is trained end-to-end under a simple diffusion objective, eliminating dependency on external speech tokenizers. Trained on a massive 1.8 million hours of bilingual corpus, our VoxCPM-0.5B model achieves state-of-the-art zero-shot TTS performance among open-source systems, demonstrating that our approach delivers expressive and stable synthesis. Besides, VoxCPM shows the capability to comprehend text to infer and generate appropriate prosody and style, delivering speech with context-aware expressiveness and natural flow. To facilitate community-driven research and development, VoxCPM is publicly accessible under Apache 2.0.

cs.SD↗

arXiv2025

A Scalable Pipeline for Enabling Non-Verbal Speech Generation and Understanding

Non-verbal Vocalizations (NVs), such as laughter and sighs, are vital for conveying emotion and intention in human speech, yet most existing speech systems neglect them, which severely compromises communicative richness and emotional intelligence. Existing methods for NVs acquisition are either costly and unscalable (relying on manual annotation/recording) or unnatural (relying on rule-based synthesis). To address these limitations, we propose a highly scalable automatic annotation framework to label non-verbal phenomena from natural speech, which is low-cost, easily extendable, and inherently diverse and natural. This framework leverages a unified detection model to accurately identify NVs in natural speech and integrates them with transcripts via temporal-semantic alignment method. Using this framework, we created and released \textbf{NonVerbalSpeech-38K}, a diverse, real-world dataset featuring 38,718 samples across 10 NV categories collected from in-the-wild media. Experimental results demonstrate that our dataset provides superior controllability for NVs generation and achieves comparable performance for NVs understanding.

cs.SD↗

arXiv2024

The Codec Language Model-based Zero-Shot Spontaneous Style TTS System for CoVoC Challenge 2024

This paper describes the zero-shot spontaneous style TTS system for the ISCSLP 2024 Conversational Voice Clone Challenge (CoVoC). We propose a LLaMA-based codec language model with a delay pattern to achieve spontaneous style voice cloning. To improve speech intelligibility, we introduce the Classifier-Free Guidance (CFG) strategy in the language model to strengthen conditional guidance on token prediction. To generate high-quality utterances, we adopt effective data preprocessing operations and fine-tune our model with selected high-quality spontaneous speech data. The official evaluations in the CoVoC constrained track show that our system achieves the best speech naturalness MOS of 3.80 and obtains considerable speech quality and speaker similarity results.

cs.SD↗

arXiv2024

Direct Readout of Nitrogen-Vacancy Hybrid-Spin Quantum Register in Diamond by Photon Arrival Time Analysis

Quantum state readout plays a pivotal role in quantum technologies, spanning applications in sensing, computation, and secure communication. In this work, we introduce a new approach for efficiently reading populations of hybrid-spin states in the nitrogen-vacancy center of diamond using a single laser pulse, which utilizes the excited state level anti-crossing mechanism at around 500 Gs. Reading spin state populations through this approach achieves the same outcome as traditional quantum state diagonal tomography but significantly reduces the experimental time by an order of magnitude while maintaining fidelity. Moreover, this approach may be extended to encompass full-state tomography, thereby obviating the requirement for a sequence of spin manipulations and mitigating errors induced by decoherence throughout the procedure.

quant-ph↗

arXiv2022

Online optimization for optical readout of a single electron spin in diamond

The nitrogen-vacancy (NV) center in diamond has been developed as a promising platform for quantum sensing, especially for magnetic field measurements in the nano-tesla range with a nanometer resolution. Optical spin readout performance has a direct effect on the signal-to-noise ratio (SNR) of experiments. In this work, we introduce an online optimization method to customize the laser waveform for readout. Both simulations and experiments reveal that our new scheme optimizes the optically detected magnetic resonance in NV center. The SNR of optical spin readout has been witnessed a 44.1% increase in experiments. In addition, we applied the scheme to the Rabi oscillation experiment, which shows an improvement of 46.0% in contrast and a reduction of 12.1% in mean deviation compared to traditional constant laser power SNR optimization. This scheme is promising to improve sensitivities for a wide range of NV-based applications in the future.

quant-ph↗

arXiv2021

Machine-learning-assisted electron-spin readout of nitrogen-vacancy center in diamond

Machine learning is a powerful tool in finding hidden data patterns for quantum information processing. Here, we introduce this method into the optical readout of electron-spin states in diamond via single-photon collection and demonstrate improved readout precision at room temperature. The traditional method of summing photon counts in a time gate loses all the timing information crudely. We find that changing the gate width can only optimize the contrast or the state variance, not both. In comparison, machine learning adaptively learns from time-resolved fluorescence data, and offers the optimal data processing model that elaborately weights each time bin to maximize the extracted information. It is shown that our method can repair the processing result from imperfect data, reducing 7% in spin readout error while optimizing the contrast. Note that these improvements only involve recording photon time traces and consume no additional experimental time, they are thus robust and free. Our machine learning method implies a wide range of applications in precision measurement and optical detection of states.

quant-ph↗