Text-to-speech (TTS) systems are conventionally optimized for intelligibility and speaker similarity. However, neither objective captures the prosodic detail that distinguishes human performances from synthetic speech, leaving a realism gap that larger models and higher-quality data can narrow but not close. In the absence of an objective that directly targets this gap, realism remains a persistent challenge in TTS. We present Audio8-TTS, a compact 0.6B-parameter model for zero-shot voice cloning that achieves state-of-the-art performance in intelligibility and speaker similarity, with particularly strong results in perceptual naturalness. To align the training signal with human perception rather than signal-level fidelity, we construct a quality ladder from widely used open-source TTS systems and train a realism critic to assess generated speech against matched human recordings of the same sentences. An identity-erasure objective discourages the critic from relying on generator fingerprints. We train the model in three stages: multilingual supervised fine-tuning; group relative policy optimization (GRPO) with word error rate (WER) and speaker similarity (SIM) rewards; and GRPO guided by the realism critic. The final stage improves perceived prosodic naturalness and speaker similarity while preserving WER performance.
Figures & tables
Figure 1: Overview of Audio8-TTS. The slow autoregressive branch predicts text tokens and the first audio codebook at the frame rate, while a shallow fast branch predicts the remaining nine codebooks in parallel. The audio decoder remains frozen.
SFT
GRPO (WER/SIM)
GRPO (realism)
CV3, error rate (%) ↓
5.41
5.01
4.86
CV3, SIM ↑
0.639
0.643
0.661
Add. languages † , error rate (%) ↓
22.36
22.53
20.77
Add. languages † , SIM ↑
0.638
0.637
0.679
Table 1: Stage-wise results on the nine CV3 languages and thirteen additional languages. Error rate is WER, except for Chinese, Japanese, and Korean, where it is CER. The same recognizer and speaker encoder are used throughout. † The additional 13 languages are outside CV3 and less represented in the training data; their results are reported separately.
Params.
Seed-TTS (en/zh)
CV3 (9)
MiniMax (24)
Model
(B)
Error ↓
SIM ↑
Error ↓
SIM ↑
Error ↓
SIM ↑
Seed-TTS [ 1 ]
∼
1.68a
0.779a
∼
∼
∼
∼
MiniMax [ 37 ]
∼
1.24a
0.738a
∼
∼
3.77a
0.766a
IndexTTS-2.5 [ 19 ]
0.81p,ℓ
1.66a
0.852a,f
∼
∼
∼
∼
Qwen3-TTS-12Hz-0.6B [ 15 ]
0.6
1.12a
∼
∼
∼
∼
∼
Qwen3-TTS-1.7B [ 15 ]
1.7
1.01a
∼
7.73h
∼
27.41h
∼
Table 2: Results on Seed-TTS, CV3, and MiniMax. Error rates (%) are macro-averaged over languages. Seed-TTS averages English WER and Chinese CER, excluding zh-hard. ∼ : unavailable. a Means computed from reported per-language results. c Results compiled in the VoxCPM repository [ 24 ] . e Our evaluation. h Boson AI’s evaluation [ 4 ] . f FunASR-based SIM. p Parameter counts computed from checkpoint tensors, excluding codec weights and buffers; otherwise, counts are taken from model releases. ℓ Excludes the separate acoustic decoder. Parameter-counting conventions differ across models.
We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch with a hybrid VQ-FSQ quantization scheme. In human evaluations conducted by native speakers, Voxtral TTS is preferred for multilingual voice cloning due to its naturalness and expressivity, achieving a 68.4% win rate over ElevenLabs Flash v2.5. We release the model weights under a CC BY-NC license.
Zero-shot text-to-speech (TTS) typically conditions on a short speech prompt, which conveys timbre but provides only limited evidence of a speaker's prosodic patterns. We formulate Long-Prompt Zero-Shot TTS, a task that uses long speech prompts at inference time to reproduce speaker-specific prosody without target-speaker fine-tuning. We introduce LP-TTS-Bench, a bilingual benchmark of 37 speakers with 63.95 prompt hours and 2,728 held-out test utterances. Beyond intelligibility and timbre similarity, LP-TTS-Bench evaluates timing, prominence, intonation, and model-based prosody metrics, enabling multidimensional evaluation of prosody reproduction. As a reference system, we develop LPTTS, whose hierarchical Long-Prompt Encoder compresses long speech prompts into a fixed number of continuous prosody vectors. Experiments show improvements in several prosodic dimensions over a short-prompt baseline, with overall gains as prompt duration increases up to approximately 10 minutes and saturation at longer durations.
Dekun Chen, Renjie Xu, Yingda Shen +2
The Chinese University of Hong Kong, Shenzhen · Amphion Technology Co., Ltd.
We present JaiTTS-v1.0, a state-of-the-art Thai voice cloning text-to-speech model built through continual training on a large Thai-centric speech corpus. The model architecture is adapted from VoxCPM, a tokenizer-free autoregressive TTS model. JaiTTS-v1.0 directly processes numerals and Thai-English code-switching, which is very common in realistic settings, without explicit text normalization. We test the models on short- and long-duration speech generation, which reflects many real-world use cases. JaiTTS-v1.0 achieves a state-of-the-art CER of 1.94%, surpassing the human ground truth of 1.98% for short-duration tasks while performing on par with human ground truth for long-duration tasks. In human judgment evaluations, our model wins 283 of 400 pairwise comparisons against commercial flagships, with only 58 losses. Our code and demo are available at https://github.com/JTS-AI-Team/JaiTTS .