From Deepfake Defense to End-to-End Waveforms: Speech & Avatar Research, Nov 10

Although audio-driven human animation has achieved impressive realism, it still lacks effective Multi-Scene Temporal Control: orchestrating background transitions at precise timestamps while preserving a coherent foreground subject. We show that this challenge stems from the diffusion transformer's self-attention mechanism, and addressing it requires two capabilities: (1) Semantic-Aware Temporal Decoupling, which isolates background attention across scenes while retaining cross-scene foreground attention for subject consistency; and (2) Foreground Motion Localization, which accurately tracks the foreground subject across latent frames, especially under large motion. To address these challenges, we introduce SceneShifter, a training-free framework for precise multi-scene control in audio-driven human animation. SceneShifter guides spatio-temporal self-attention by suppressing cross-scene attention among background tokens while preserving cross-scene attention among foreground tokens. To support this guidance under large motion, SceneShifter extracts dynamic foreground masks from the self-attention layers and heads that best capture subject trajectories. We further introduce SceneShifterBench, a benchmark designed to evaluate scene-timing accuracy, foreground preservation, and visual quality under large motion, multiple subjects. Experiments show that SceneShifter achieves frame-level scene timing, strong subject preservation, and high visual quality, outperforming existing baselines. Code is available at https://github.com/JamesLong199/SceneShifter.

Talking-head generation models, which synthesize realistic facial animations from audio, are increasingly vulnerable to misuse in multimodal deepfake scenarios. Protecting such systems remains challenging, as talking-head models fundamentally rely on precise audio-visual alignment, while effective audio protection methods remain largely underexplored. In this work, we propose MUTE, an audio protection framework that uncouples the underlying audio-visual alignment through a multi-level strategy. Specifically, MUTE combines (1) Alignment-level Disruption, which directly perturbs cross-modal attention to disrupt structural alignment, and (2) Representation-level Degradation, which perturbs temporal audio embeddings to degrade their representations. To improve robustness, perturbations are constrained in the STFT domain to high-energy regions, making them resistant to post-processing, denoising, and adaptive attacks. An optional speaker-level objective further mitigates potential bypass via TTS-based resynthesis. Extensive experiments demonstrate that MUTE consistently degrades lip synchronization across both white-box and black-box talking-head models while preserving perceptual audio quality. The proposed method remains effective under real-world transformations and can be combined with image-based approaches to provide stronger multimodal protection. The source code and project page are available at https://github.com/jindl465/MUTE.

We present M5-TTS, a compact text-to-speech system supporting multilingual and Chinese multidialect synthesis, zero-shot voice cloning, speaker-specific multilingual generation, and CPU streaming. Its 50M-parameter language model, excluding the speech codec, combines a temporal backbone with a local multi-token prediction module over 16 codebooks. GRPO post-training combines global speaker-similarity and pause feedback with a token-level CER process advantage. Character-level error comparisons provide local feedback, with negative credit propagated backward within a clause and projected onto audio-token positions. M5-TTS achieves 1.37% CER on Seed-ZH and 2.91% WER on Seed-EN, corresponding to relative error reductions of 62.2% and 35.3% over MOSS-TTS-Nano with half the language-model parameters. It also achieves lower content error rates in five of six evaluated MiniMax languages, while difficult Chinese inputs remain a limitation. On four Intel Xeon Gold 6530 cores, it achieves a real-time factor of 0.653 and a warm first-chunk latency of 225ms. A 10-listener evaluation gives M5-TTS higher speaker-similarity and intelligibility ratings than MOSS-TTS-Nano, with a lower naturalness rating. We will release training, post-training, and inference code, together with voice-cloning and speaker-specific checkpoints, to support reproducible research on compact speech generation for on-device voice interaction.

This work proposes WaveGen, a fully end-to-end diffusion-based text-to-waveform framework based on Spectral Forcing, which jointly learns an internal spectral trajectory and the target waveform trajectory within a single model, enabling direct end-to-end text-to-waveform generation. Unlike recent TTS pipelines, WaveGen does not rely on separately trained or externally decoded intermediate representations, such as neural audio codecs, Mel-spectrogram vocoders, or VAE-based latent models. In addition, WaveGen performs internal text-speech alignment within a single model, eliminating the need for external alignment modules such as duration predictors. To preserve a truly end-to-end formulation, our framework further avoids dependence on self-supervised representations, such as BERT-based text embeddings and Wav2Vec 2.0-based semantic representations, as well as semantic distillation methods for training optimization. To achieve this, we carefully design Spectral Forcing, a model architecture with disentangled diffusion heads, and a training objective based on a multi-scale spectral $v$-loss. We further validate the scalability of end-to-end TTS models by scaling the model size from 0.1B to 0.6B parameters. Using a single architecture, open-source data, and one-stage training, without external modules such as duration predictor or REPA, WaveGen demonstrates fully end-to-end waveform generation achieving state-of-the-art performance on the Emilia training dataset. In particular, WaveGen achieves a WER of 1.47 and a SIM of 0.75 on the Seed-en benchmark. Source code and checkpoints are available at https://github.com/sh-lee-prml/WaveGen

Text-to-speech systems built from a language model and a flow-matching mel decoder now run the decoder with two to four Euler steps instead of the official ten or more. At two steps the word error rate (WER) is unchanged, but speaker similarity (SIM) drops by one to six points. This paper shows that most of this loss comes from classifier-free guidance on the first step. At pure noise the network can only return the conditional mean, so guidance shifts every trajectory by nearly the same vector. Later steps do not remove this shift, and its size grows with the length of the first step, so it is negligible with ten or more steps and accounts for most of the loss with two. Two training-free samplers are derived from this analysis. First-step unguided sampling (FSU) leaves the first step unguided and recovers $84$--$91\%$ of the two-step SIM loss on CosyVoice~2/3 with no significant WER increase. FSU-Mid replaces the two Euler steps by one midpoint step with an unguided first evaluation, recovers $77$--$85\%$ on IndexTTS-2 and matches the ten-step teacher on CosyVoice~2 and GLM-TTS. Both use three network function evaluations, one fewer than the official two-step sampler, and FSU-Mid gives the highest mean recovery among eight published guidance rules re-run at two steps on four decoders.

We present BARS (Beat-Adaptive Rap Synthesis), a non-autoregressive rapping voice generation model based on Flow Matching, conditioned on a beat track (instrumental backing), lyrics, and a short reference voice for zero-shot timbre control. Unlike prior autoregressive rap synthesis, where rap pace and output duration are strictly determined by the beat, BARS enables independent control over the rap tempo, allowing flexible-length generation without explicit duration prediction or phoneme-level alignment. We encode lyrics using a pretrained ByT5 encoder with a lightweight Transformer adapter trained jointly with the model, without relying on speech-pretrained text encoders, enabling natural handling of non-standard orthography (e.g., slang, abbreviations) and numerals common in rap lyrics. To address alignment instabilities, such as token skipping or repetition, under strong beat conditioning, we propose Asymmetric Scaled RoPE for cross-attention, which introduces a positional structure to facilitate sequential lyric-audio alignment learning. Furthermore, the model supports partial re-generation via span masking, enabling lyric-level editing while preserving the surrounding vocal style. Experiments demonstrate that the proposed model achieves high lyric fidelity, natural rap flow that is rhythmically coherent with the beat, and robust tempo control across a wide range of rapping styles and speeds.

As modern text-to-speech systems become increasingly intelligible and natural, evaluation is shifting from detecting gross synthesis failures to identifying subtle departures from lifelike human delivery. Because large-scale listening tests are costly and slow, speech reward models can serve as scalable automatic judges for comparing generated utterances, curating data, and supporting iterative TTS development. Existing speech reward models can predict scores or pairwise preferences and may generate natural-language rationales, but their supervision is typically defined at the utterance or pair level. The connection between local acoustic phenomena and the final preference therefore remains implicit. We introduce REAL, a framework for Reward modeling through Evidence-linked Analysis of Lifelikeness in generated speech. Its central idea is to supervise not only the final pairwise preference but also a structured account of its supporting observations, linking local audio spans to perceptual phenomena, affected dimensions, and the overall judgment. Specifically, REAL-Jury constructs these evidence-linked reports through timestamp alignment, multi-model analysis, and dispute-oriented deliberation. REAL-Judge learns to generate them from paired audio inputs. REAL-Eval assesses preference prediction in two tiers, from salient perceptual differences to subtle distinctions among strong TTS systems. Instantiated with MiMo-Audio, REAL-Judge improves pairwise preference accuracy over its backbone by 16.01 and 11.41 percentage points on the two tiers, respectively, and outperforms the open-source speech reward models SpeechJudge and UniSRM on both tiers.

Cascaded spoken dialogue systems comprise perception-and-comprehension, decision-making, and response-generation modules. We study iterative optimization of the first and last modules. Their reciprocal speech-content interfaces enable one-sided data to provide cross-module supervision: acoustic observations, such as waveforms or feature sequences, can be pseudo-labeled to train generation, whereas content-only samples can be rendered into speech to train comprehension. Because the reliable teacher-student direction varies across domains, we propose a two-stage, capability-guided pipeline for iterative online optimization. Stage 1 compares both modules through a shared intelligibility probe and obtains a domain-wise capability prior. Stage 2 activates the compatible one-sided-data route, constructs and filters supervision with the current teacher, updates the selected student, and refreshes the relevant checkpoints across cycles. We instantiate comprehension with Whisper large-v3 and generation with CosyVoice2. After only 200 steps for Whisper and 600 for CosyVoice2 of our iterative online optimization, we observe that, on the Chinese MAGICDATA-RAMC dataset, Whisper reduces CER from 14.18% to 10.77%, while CosyVoice2 reduces it from 4.78% to 4.04%. On the Japanese Common Voice dataset, Whisper reduces CER from 15.23% to 15.10%, while CosyVoice2 reduces it from 14.85% to 11.18%. These results demonstrate the feasibility of capability-guided iterative optimization of complementary spoken-dialogue modules.