Voice Control, Timbre Localization, and Adaptive Codecs in Speech Synthesis

Voice style conversion (VSC) aims to transform a source utterance to match the timbre, accent, and emotion of a target voice while preserving linguistic content. StyleStream 1.0 introduced the first streamable zero-shot VSC system with state-of-the-art conversion quality, but remained limited in two ways: approximately 1s end-to-end latency on consumer hardware, and the lack of a text-based interface for designing or editing target voices. We present StyleStream 2.0, a fast and controllable streaming VSC framework that addresses these limitations. To reduce latency, we train a few-step pixel MeanFlow model and further fine-tune it with autoregressive feedback inspired by Self Forcing, improving robustness under streaming inference. This reduces end-to-end latency to 520~ms on a consumer GPU, a 2x speedup over StyleStream 1.0, while maintaining conversion quality. To enable controllable style manipulation, we remove the mel-spectrogram context and instead route target style information through a compact style embedding. In this embedding space, we train a unified text-conditioned flow matching model that supports both text-based voice design, which maps natural language prompts to voice styles, and instruction-based voice editing, which modifies a specific attribute of a source utterance while preserving the rest. Experiments show that StyleStream 2.0 achieves the strongest target style fidelity on VSC, voice design, and voice editing, while remaining competitive on intelligibility and non-edited attribute preservation.

Zero-shot cross-lingual voice cloning often carries the reference speaker's accent into the target language. We propose closed-loop dual-reference guidance, which pairs the original reference with a reusable native anchor constructed by voice conversion (VC). The references initialize separate language-model (LM) states that guide a shared acoustic trajectory; each completed patch then updates both states, keeping their generated histories aligned with the emitted speech. Anchor construction runs once per speaker, and online synthesis uses a frozen TTS model without per-utterance VC. On 20 held-out French speakers, the fixed setting $λ^{\!*}=0.33$ reduces residual French LID mass by 73\% versus direct cloning, at an ECAPA similarity cost of only $0.042$; relative to anchor-only cloning, it recovers $0.155$. The method achieves the lowest mean WER ($0.009$) and highest mean GOP ($4.750$) among the held-out test arms. In blinded preference tests, native-English listeners favor it over direct cloning on pronunciation naturalness, and over both anchor-only cloning and a TTS$\to$VC cascade on speaker identity. Online RTF is approximately 26\% lower than the cascade's ($0.59$ vs.\ $0.80$), and the same setting reduces Spanish LID mass by 56\% without re-tuning.

This paper addresses fine-grained prosody control for large-language-model (LLM) based text-to-speech (TTS) synthesis of Mandarin Chinese, motivated by multimedia content-creation scenarios---film dubbing, audiobook production, and interactive voice interfaces---in which prosodic adjustments must be placed on specific characters rather than applied uniformly to whole utterances. We present a data-driven framework in which fine-grained prosodic markers (pitch, duration, energy, and pause) are automatically extracted from raw audio at scale and embedded into the text stream as out-of-vocabulary Unicode symbols. The LLM of an existing TTS system is then fine-tuned to interpret these markers while its architecture remains unchanged, yielding character-level temporal control that complements existing utterance-level style control. An optional text-LLM front-end can translate natural-language instructions into marker sequences. Experiments on 2.15M Mandarin utterances (5,858 hours) with CosyVoice2 as the backbone demonstrate reliable control of character duration, pitch, pause duration, and global speaking rate, with perceptual quality essentially preserved. A cross-tokenizer analysis further indicates that the ASR-optimized tokenizer of the backbone attenuates prosodic variation relative to reconstruction-oriented alternatives---an observation we report as hypothesis-generating evidence for prosody-sensitive system design.

High linear decodability does not entail causal controllability in continuous-latent text-to-speech (TTS). In VoxCPM2, position-matched probes and equal-norm interventions across 30 speakers and 60 directed dyads expose this separation: reference-audio states at the final residual acoustic language model layer (RALM L7) retain 0.970 probe accuracy, yet aligned edits yield 0/60 target-dominant dyads. Edits at early reference-audio positions produce the opposite pattern. At L0, pair-aligned updates increase target similarity by 0.326 and achieve target dominance in all 60 dyads, against at most 1/60 under any equal-norm direction or position control. Independent speaker encoders, retrieval against held-out speaker prototypes, and double-blind listening confirm that the shifted voice is heard as the intended target. Cross-architecture interventions in dots.tts distinguish backbone leverage from direct acoustic control: backbone edits displace the source without target attraction, while x-vector ablation reduces source similarity from 0.849 to 0.006. These findings identify two requirements for training-free speaker control: geometric specificity and a computational route that carries the intended speaker condition into acoustic synthesis.

Neural speech codecs provide discrete tokens for speech generation. Lower frame rates shorten token sequences, but each token must then carry more linguistic and acoustic information. Variable-frame-rate codecs address this trade-off by adapting token allocation to speech content, but threshold-based allocation lets content determine the token count, so the sequence length a speech generator must produce varies across utterances and datasets at a fixed setting. We present RubatoCodec, a variable-frame-rate speech codec built on X-Codec-2 with an exact utterance-level event budget and streaming decoding. Its Adaptive Firing mechanism uses changes in codec features and waveform spectra to divide speech into variable-duration events without a learned boundary predictor. Discrete codes and an explicit duration represent each event, allowing a speech generator to predict and decode one event at a time. Controlled experiments show that adaptive allocation outperforms fixed-rate coding in reconstruction fidelity and intelligibility at low event rates. A model trained on large-scale multilingual speech achieves competitive reconstruction quality across English, Japanese, and Mandarin, with the best intelligibility preservation among the compared codecs on the evaluated English and Japanese benchmarks at comparable bitrates. In a matched text-to-speech comparison, the adaptive representation yields lower WER and higher UTMOS and UTMOSv2 with comparable speaker similarity. The speech generator also learns the codec's utterance-level event budget without an explicit generation-time rate constraint. These results demonstrate that budgeted adaptive speech representations support both high-quality reconstruction and streaming text-to-speech generation. Code and weights are available at https://github.com/third-intelligence/RubatoCodec and https://huggingface.co/third-intelligence/RubatoCodec.

Timbre similarity is a key evaluation target for generative speech systems, including voice conversion, speaker-cloning text-to-speech, and zero-shot speech models. When the generated voice does not match the target speaker, the problem often reflects imperfect separation of speaker-related timbre from content, making it important to identify where timbre information resides and which representations still carry it. Using speaker identity as a measurable proxy, we probe frequency bands, phase/magnitude structure, WORLD source/filter decomposition, voiced/unvoiced regions, self-supervised learning (SSL) layers, EnCodec/HuBERT tokens, and a 45-listener test across Mandarin, English, and Japanese speech/singing data. Speaker information is distributed: 1--8 kHz magnitude and vocal-tract spectral-envelope structure dominate perception, above-8 kHz bands remain machine-discriminative, shallow SSL layers expose compact speaker information, and token leakage depends on tokenizer/layer choice. These findings indicate where generation systems should condition on or transform speaker information, and where content, codec, and token pipelines should audit residual speaker leakage.