NaturalSpeech 3: Zero-Shot Speech Synthesis with Factorized Codec and Diffusion Models
read the original abstract
While recent large-scale text-to-speech (TTS) models have achieved significant progress, they still fall short in speech quality, similarity, and prosody. Considering speech intricately encompasses various attributes (e.g., content, prosody, timbre, and acoustic details) that pose significant challenges for generation, a natural idea is to factorize speech into individual subspaces representing different attributes and generate them individually. Motivated by it, we propose NaturalSpeech 3, a TTS system with novel factorized diffusion models to generate natural speech in a zero-shot way. Specifically, 1) we design a neural codec with factorized vector quantization (FVQ) to disentangle speech waveform into subspaces of content, prosody, timbre, and acoustic details; 2) we propose a factorized diffusion model to generate attributes in each subspace following its corresponding prompt. With this factorization design, NaturalSpeech 3 can effectively and efficiently model intricate speech with disentangled subspaces in a divide-and-conquer way. Experiments show that NaturalSpeech 3 outperforms the state-of-the-art TTS systems on quality, similarity, prosody, and intelligibility, and achieves on-par quality with human recordings. Furthermore, we achieve better performance by scaling to 1B parameters and 200K hours of training data.
This paper has not been read by Pith yet.
Forward citations
Cited by 32 Pith papers
-
WavTTS: Towards High-Quality Zero-Shot TTS via Direct Raw Waveform Modeling
WavTTS is the first raw-waveform diffusion TTS model using DiT flow matching and multi-scale mel supervision that approaches SOTA latent zero-shot performance while beating prior end-to-end models.
-
FlexiSLM: A Dynamic and Controllable Frame Rate Spoken Language Model
FlexiSLM is the first spoken language model supporting dynamic and controllable frame rates on speech input and output, outperforming fixed-rate 7B models at high quality and enabling faster inference at lower rates l...
-
AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation
AudioCALM presents a continuous autoregressive framework with flow-matching prediction and A-MoME architecture that unifies speech, sound, and music generation while matching modality-specific state-of-the-art performance.
-
Bagpiper-TTS: Natural Language Guided Universal Speech Synthesis
Bagpiper-TTS uses natural language prompts and intent reasoning to derive rich captions that guide a single model for universal speech synthesis across classical TTS, multi-talker, singing, and role-play tasks.
-
AffectCodec: Emotion-Preserving Neural Speech Codec for Expressive Speech Modeling
AffectCodec is an emotion-guided neural speech codec that preserves emotional cues during quantization while maintaining semantic fidelity and prosodic naturalness.
-
PairAlign: A Framework for Sequence Tokenization via Self-Alignment with Applications to Audio Tokenization
PairAlign learns compact variable-length token sequences for audio via self-alignment on paired content-preserving views, achieving 55% fewer archive tokens than VQ while preserving edit-distance retrieval at 12.71 tokens/s.
-
PairAlign: A Framework for Sequence Tokenization via Self-Alignment with Applications to Audio Tokenization
PairAlign learns compact audio token sequences via self-alignment of paired content views using an autoregressive decoder, achieving strong cross-view consistency and edit-distance preservation while reducing token co...
-
Minimizing Modality Gap from the Input Side: Your Speech LLM Can Be a Prosody-Aware Text LLM
TextPro-SLM minimizes the speech-text modality gap from the input side via a prosody-aware unified encoder, delivering the lowest gap and strong performance at 3B/7B scales with only ~1000 hours of audio.
-
JUST-DUB-IT: Video Dubbing via Joint Audio-Visual Diffusion
JUST-DUB-IT adapts a joint audio-visual diffusion model via LoRA to generate high-quality dubbed videos with translated audio and lip-synced facial motion.
-
Moshi: a speech-text foundation model for real-time dialogue
Moshi is the first real-time full-duplex spoken large language model that casts dialogue as speech-to-speech generation using parallel audio streams and an inner monologue of time-aligned text tokens.
-
FineCombo-TTS: Collaborative and Precise Controllable Speech Synthesis Using Text Descriptions and Reference Speech
FineCombo-TTS learns a unified acoustic representation with a CFM-based Speech Variance Predictor for flexible precise TTS control from reference audio and text descriptions, supported by the new FineEdit paired dataset.
-
DDPO-VC: Speaker De-Identification via Diffusion Denoising Policy Optimization
DDPO-VC applies diffusion denoising policy optimization with dual-teacher rewards to improve speaker de-identification while preserving cognitive utility on dementia speech benchmarks.
-
Self-Guidance: Enhancing Neural Codecs via Decoder Manifold Alignment
Self-guidance adds a lightweight feature-mapping loss to align decoder manifolds in VQ-VAE speech codecs, raising reconstruction metrics and allowing 4x codebook reduction with no fidelity loss.
-
EmoInstruct-TTS: Dual-Path Instruction-Guided Emotional Speech Synthesis
EmoInstruct-TTS uses Emotion2embed and an Instruction-Conditioned Emotion Flow Model (ICE-Flow) to generate acoustically grounded emotion representations from free-form instructions and integrate them into an LLM-base...
-
Can We Hear from Events? Generating Speech from Event Camera
EventSpeech is a text-conditioned neural framework that uses neuromorphic event cameras and a new EVT-SPK benchmark to generate expressive speech, claiming to outperform RGB baselines by preserving fine-grained emotio...
-
SemaVoice: Semantic-Aware Continuous Autoregressive Speech Synthesis
SemaVoice adds SFM-guided alignment to refine continuous speech representations in autoregressive TTS, reporting 1.71% English WER on Seed-TTS and competitiveness with open-source SOTA.
-
Text-To-Speech with Chain-of-Details: modeling temporal dynamics in speech generation
Chain-of-Details (CoD) is a cascaded TTS method that explicitly models temporal coarse-to-fine dynamics with a shared decoder, achieving competitive performance using significantly fewer parameters.
-
MimicLM: Zero-Shot Voice Imitation through Autoregressive Modeling of Pseudo-Parallel Speech Corpora
MimicLM achieves better naturalness in zero-shot voice imitation by autoregressively modeling pseudo-parallel data with synthetic sources and real targets, plus interleaved text-audio guidance and preference alignment.
-
Qwen3-TTS Technical Report
Qwen3-TTS delivers state-of-the-art multilingual TTS performance with 3-second voice cloning, description control, and ultra-low-latency streaming via dual tokenizers and a dual-track LM architecture trained on over 5...
-
JAM-Flow: Joint Audio-Motion Synthesis with Flow Matching
JAM-Flow introduces a unified flow-matching model with a Multi-Modal Diffusion Transformer that jointly synthesizes facial motion and speech from text, audio, or motion inputs.
-
AugCodec: A Low-Bitrate Disentangled Neural Speech Codec via Data Augmentation
AugCodec disentangles speech into semantic, speaker, and prosody tokens via tailored data augmentations, achieving 12.5 Hz operation with three streams and outperforming prior codecs on LibriSpeech reconstruction and ...
-
Which Speech Representation Better Matches Text-Native Reasoning? A Study of Speech-Text Alignment on Frame Rate and Representation
Empirical sweep finds 4.17 Hz frame rate plus intermediate-layer alignment optimal for speech QA under frozen text LLM backbone.
-
UniVoice: A Unified Model for Speech and Singing Voice Generation
UniVoice is a conditional flow matching model with a Diffusion Transformer backbone that unifies TTS and SVS via modality-specific encoders and a null melody token for speech, achieving 5.26% speech PER and 16.22% sin...
-
Minimizing Modality Gap from the Input Side: Your Speech LLM Can Be a Prosody-Aware Text LLM
TextPro-SLM reduces the speech-text modality gap by feeding an LLM backbone with synchronized text tokens and prosody embeddings from WhisperPro, achieving lowest gap scores at 3B/7B scales with roughly 1,000 hours of audio.
-
Controllable Singing Style Conversion with Boundary-Aware Information Bottleneck
A singing voice conversion system with boundary-aware information bottleneck and high-frequency augmentation achieves the best naturalness in SVCC2025 subjective tests while using less extra data than competitors.
-
F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching
F5-TTS generates natural speech from text via flow matching on DiT with simple text padding, ConvNeXt refinement, and sway sampling, trained on 100K hours multilingual data.
-
MAVIN: Multi-Shot Audio-Visual Generation with Narrative Control
MAVIN proposes boundary-aware attention, ID-aware propagation, a multi-agent scripting pipeline, and the MAVINSet dataset as the first framework for multi-shot audio-visual generation with narrative control, claiming ...
-
Multimodal Group Emotion Recognition In-the-Wild Towards a Privacy-Safe Non-Individual Approach
Proposes cross-attention audio-video fusion and VE-MD latent-space models for group emotion recognition that avoid individual cues and report competitive performance via ablation studies on synthetic and real data.
-
SPARCLE: SPeaker-aware Aligned Representations via Contrastive Language Embeddings
SPARCLE builds speaker-aware grapheme representations by contrastively aligning characters with Wav2Vec2 acoustic embeddings conditioned on speaker identity, replacing G2P for TTS and halving WER in low-resource cases.
-
ATRIE: Adaptive Tuning for Robust Inference and Emotion in Persona-Driven Speech Synthesis
ATRIE disentangles timbre and prosody in a Persona-Prosody Dual-Track model distilled from a large LLM to achieve strong identity preservation (EER 0.04) and emotional speech synthesis with SOTA results on an extended...
-
Multimodal Large Language Model-Enabled Video Translation: A Role-Oriented Survey
The paper offers the first focused review of MLLM-based video translation organized by a three-role taxonomy of Semantic Reasoner, Expressive Performer, and Visual Synthesizer, plus open challenges.
-
Voice Mapping of Text-to-Speech Systems: A Metric-Based Approach for Voice Quality Assessment
Voice range indicates TTS model capability with VITS highest, Glow-TTS best at soft phonation, and CPPs of 7-8 dB marking natural quality while values over 10 dB sound robotic.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.