Higgs Audio V2

Higgs Audio V2 is an open-source audio foundation model developed by Boson AI, pretrained on more than 10 million hours of audio data. The model achieves highly expressive audio generation without post-training or fine-tuning, reaching industry-leading levels in emotional expressiveness and multi-speaker dialogue generation.

Key Features

• Zero-shot voice cloning: Generate speech with a similar timbre based on reference audio
• Multi-speaker dialogue generation: Supports natural, fluid multi-person conversations with automatically matched energy and emotion
• Multilingual audio generation: Text-to-speech in multiple languages
• Automatic prosody adaptation: Automatically adjusts tone and rhythm during narration
• Emotional expressiveness: Achieved a 75.7% win rate in the emotion category on EmergentTTS-Eval, surpassing GPT-4o-mini-tts
• Music and speech fusion: Can generate background music and speech content at the same time

Use Cases

In content creation, creators can use it to generate podcasts, audiobooks, or video voiceovers; in multimedia production, it can quickly generate multi-character dialogue audio; for voice cloning, developers can build personalized speech synthesis from a small amount of reference audio.

Unique Advantages

Higgs Audio V2 uses an innovative DualFFN architecture built on top of Llama-3.2-3B, with 5.8B total parameters. Audio quality has been upgraded from 16kHz to 24kHz, with excellent performance in speaker similarity and overall naturalness. As the first open-source model with genuine 'awareness' of emotional expression, it excels at controlling intonation, timing, and tone.