[Paper 리뷰] DisCo-Speech: Controllable Zero-Shot Speech Generation with a Disentangled Speech Codec
DisCo-Speech: Controllable Zero-Shot Speech Generation with a Disentangled Speech Codec기존 codec은 timbre, prosody의 entanglement로 인해 independent control이 어려움DisCo-SpeechParallel encoder와 hybrid loss를 사용하여 speech를 content, prosody, timbre의 tri-factor로 disentangleUnified content-prosody token을 구성해 disentanglement-reconstruction trade-off를 balance논문 (ACL 2026) : Paper Link1. IntroductionCodec-based L..
[Paper 리뷰] Emo-LiPO: Listwise Preference Optimization for Fine-Grained Emotion Intensity Control in LLM-based Text-to-Speech
Emo-LiPO: Listwise Preference Optimization for Fine-Grained Emotion Intensity Control in LLM-based Text-to-SpeechLarge Language Model-based Text-to-Speech는 fine-grained emotion intensity control의 한계가 있음Emo-LiPOFixed transcript 하에서 각 emotion에 대한 global intensity ordering을 modeling추가적으로 multi-speaker dataset인 ESD-Plus를 구축논문 (IJCAI 2026) : Paper Link1. IntroductionLarge Language Model (LLM)-based T..
[Paper 리뷰] CoCoEmo: Composable and Controllable Human-Like Emotional TTS via Activation Steering
CoCoEmo: Composable and Controllable Human-Like Emotional TTS via Activation Steering대부분의 text-to-speech system은 single utterance-level emotion을 enforce 함CoCoEmoActivation steering에 대한 multi-rater evaluation protocol을 도입Human-like emotional speech를 위한 lightweight steering approach를 적용논문 (ICML 2026) : Paper Link1. IntroductionNatural speech는 inherently complex 하고 multiple concurrent, conflicting ..
[Paper 리뷰] EmoShift: Lightweight Activation Steering for Enhanced Emotion-Aware Speech Synthesis
EmoShift: Lightweight Activation Steering for Enhanced Emotion-Aware Speech SynthesisLarge Language Model은 emotion-specific latent characteristic을 modeling 하는데 한계가 있음EmoShiftOutput embedding space에서 각 target emotion에 대한 steering vector를 학습해당 EmoSteer layer를 incorporate하여 lightweight activation-steering framework를 구성논문 (ICASSP 2026) : Paper Link1. IntroductionText-to-Speech (TTS)에서 emotion contro..
[Paper 리뷰] VibeVoice: Expressive Podcast Generation with Next-Token Diffusion
VibeVoice: Expressive Podcast Generation with Next-Token DiffusionPodcast와 같은 long-form, multi-speaker conversational audio를 생성하기 위해서는 Text-to-Speech system에서 scalability, speaker consistency, natural turn-taking를 보장할 수 있어야 함VibeVoice7.5 ultra-low frame rate의 continuous speech tokenizer를 활용해 long sequence efficiency를 개선추가적으로 next-token diffusion framework를 통해 expressive podcast generation을 지원논문 ..
[Paper 리뷰] VoxCPM: Hierarchical Semantic-Acoustic Modeling via Semi-Discrete Residual Representations for Expressive End-to-End Speech Synthesis
VoxCPM: Hierarchical Semantic-Acoustic Modeling via Semi-Discrete Residual Representations for Expressive End-to-End Speech SynthesisSpeech tokenizer 기반의 multi-stage speech synthesis는 semantic-acoustic divide로 인한 trade-off가 존재함VoxCPMSemi-discrete residual representation 기반의 hierarchical semantic-acoustic modeling을 적용추가적으로 natural specialization을 위한 differentiable quantization bottleneck을 도입논문 (I..