Zyphra/ZONOS2

Zonos2 is a leading open-weight text-to-speech MoE.

What it solves

ZONOS2 is a high-fidelity text-to-speech (TTS) system designed to provide naturalistic voice cloning and expressive speech synthesis with low latency. It addresses the challenge of creating human-like audio that maintains speaker identity while allowing for precise control over emotion, speed, and quality.

How it works

ZONOS2 uses a Mixture-of-Experts (MoE) backbone to generate DAC tokens. During inference, it processes normalized UTF-8 bytes and ECAPA-TDNN speaker embeddings to synthesize speech. The system supports a wide range of languages across three tiers of support and provides multiple ways to run the model, including a high-performance inference server based on Mini-SGLang and a ggml implementation for CPU and cross-platform use.

Who it’s for

This project is for developers and researchers who need high-quality, multilingual TTS with advanced voice cloning capabilities and the ability to fine-tune emotional prosody in real-time applications.

Highlights

  • Massive Training Set: Trained on over 6 million hours of varied multilingual speech.
  • Advanced Voice Cloning: High-fidelity cloning from reference audio files.
  • Emotion Control: Ability to nudge voices toward specific emotions (happy, sad, angry, surprised) or along valence/arousal axes using additive direction vectors without changing speaker identity.
  • Extensive Language Support: Supports dozens of languages across three tiers, including English, Mandarin, Japanese, and many others.
  • Flexible Deployment: Offers a Python API for offline inference, a dedicated TTS server with a Web UI, and an OpenAI-compatible API endpoint.

Related

  • Project
    fishaudio/fish-speech

    A state-of-the-art multilingual text-to-speech system that uses a Dual-AR architecture to provide realistic voice cloning and fine-grained emotional control via natural language tags.

  • Project
    index-tts/index-tts

    A zero-shot text-to-speech system that clones voices from a single audio clip with fine-grained control over emotion, speed, and pronunciation across five languages.

  • Project
    Audio8-AI/Audio8_TTS

    A compact 0.6B-parameter multilingual text-to-speech model featuring zero-shot voice cloning and efficient deployment options for CPU and GPU.

  • Project
    neuphonic/neutts

    A collection of open-source, on-device text-to-speech models that enable natural-sounding speech and instant voice cloning locally on mobile and embedded devices.

  • Project
    pnnbao97/VieNeu-TTS

    A high-fidelity Vietnamese text-to-speech system featuring bilingual support, instant voice cloning, and optimized on-device inference for CPU and GPU.