---
title: voxtream2
canonical_url: "https://www.modelscope.cn/models/AI-ModelScope/voxtream2"
md_url: "https://www.modelscope.cn/models/AI-ModelScope/voxtream2.md"
repository: AI-ModelScope/voxtream2
last_updated: 2026-10-14
license: cc-by-nc-4.0
pipeline_tag: text-to-speech
tasks:
  - text-to-speech
parameters: 462.9M
tensor_type:
  - F32
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 35
stars: 0
tags:
  - text-to-speech
  - zero-shot
  - streaming
---

# voxtream2

> voxtream2 - AI-ModelScope 在 ModelScope 开源的模型。Model Card for VoXtream2

AI-ModelScope/voxtream2 是 ModelScope 魔搭社区上的 462.9M 参数text-to-speech模型，采用 cc-by-nc-4.0 许可。

- **Repository**: AI-ModelScope/voxtream2
- **License**: cc-by-nc-4.0
- **Tasks**: text-to-speech
- **Parameters**: 462.9M
- **Tags**: text-to-speech, zero-shot, streaming
- **Downloads**: 35
- **Stars**: 0
- **Last updated**: 2026-10-14

Source: https://www.modelscope.cn/models/AI-ModelScope/voxtream2

---

# Model Card for VoXtream2

🚨 **Update**: New version trained with an unfrozen Depth Transformer with improved objective metrics was [released](https://huggingface.co/herimor/voxtream2.1).

VoXtream2 is a zero-shot full-stream TTS model with dynamic speaking-rate control that can be updated mid-utterance on the fly. It was introduced in the paper [VoXtream2: Full-stream TTS with dynamic speaking rate control](https://huggingface.co/papers/2603.13518).

**Developed by:** Nikita Torgashov, Gustav Eje Henter, Gabriel Skantze

### Key features

- **Dynamic speed control**: Distribution matching and Classifier-free guidance allow for a fine-grained speaking rate control, which can be adjusted as the model generates speech.
- **Streaming performance**: Works **4x** times faster than real-time and achieves **74 ms** first packet latency in a full-stream on a consumer GPU.
- **Translingual capability**: Prompt text masking enables support of acoustic prompts in any language.

### Model Sources 

- **Repository:** [https://github.com/herimor/voxtream](https://github.com/herimor/voxtream) 
- **Paper:** [https://huggingface.co/papers/2603.13518](https://huggingface.co/papers/2603.13518) 
- **Demo Page:** [https://herimor.github.io/voxtream2](https://herimor.github.io/voxtream2)
- **Live Demo:** [https://huggingface.co/spaces/herimor/voxtream2](https://huggingface.co/spaces/herimor/voxtream2)

## Get started

### Installation

### eSpeak NG phonemizer

```bash
# For Debian-like distribution (e.g. Ubuntu, Mint, etc.)
apt-get install espeak-ng
# For RedHat-like distribution (e.g. CentOS, Fedora, etc.) 
yum install espeak-ng
# For MacOS
brew install espeak-ng
```

### Pip package

```bash
pip install "voxtream>=0.2"
```

### Usage

* Prompt audio: a file containing 3-10 seconds of the target voice. The maximum supported length is 20 seconds (longer audio will be trimmed).
* Text: What you want the model to say. The maximum supported length is 1000 characters (longer text will be trimmed).
* Speaking rate (optional): target speaking rate in syllables per second.

#### Output streaming
```bash
voxtream \
    --prompt-audio assets/audio/english_male.wav \
    --text "In general, however, some method is then needed to evaluate each approximation." \
    --output "output_stream.wav"
```

#### Full streaming (slow speech, 2 syllables per second)
```bash
voxtream \
    --prompt-audio assets/audio/english_female.wav \
    --text "Staff do not always do enough to prevent violence." \
    --output "full_stream_2sps.wav" \
    --full-stream \
    --spk-rate 2.0
```

* Note: Initial run may take some time to download model weights and warmup model graph.

### Out-of-Scope Use

Any organization or individual is prohibited from using any technology mentioned in this paper to generate someone's speech without his/her consent, including but not limited to government leaders, political figures, and celebrities. If you do not comply with this item, you could be in violation of copyright laws.

## Training Data

The model was trained on [Emilia](https://huggingface.co/datasets/amphion/Emilia-Dataset) and [HiFiTTS2](https://huggingface.co/datasets/nvidia/hifitts-2) datasets. You can download preprocessed dataset [here](https://huggingface.co/datasets/herimor/voxtream2-train). For more details, please check our paper. 

## Citation 

```bibtex
@inproceedings{torgashov2026voxtream,
  title={Vo{X}tream: Full-Stream Text-to-Speech with Extremely Low Latency},
  author={Torgashov, Nikita and Henter, Gustav Eje and Skantze, Gabriel},
  booktitle={Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year={2026},
  note={to appear},
  url={https://arxiv.org/abs/2509.15969}
}

@article{torgashov2026voxtream2,
  author    = {Torgashov, Nikita and Henter, Gustav Eje and Skantze, Gabriel},
  title     = {Vo{X}tream2: Full-stream TTS with dynamic speaking rate control},
  journal   = {arXiv:2603.13518},
  year      = {2026}
}
```