MARS5 TTSOpen-source, insanely prosodic text-to-speech model

MARS5 TTS: Transforming Text-to-Speech with Advanced Prosody

Introduction

MARS5 TTS by CAMB.AI is a cutting-edge text-to-speech model designed to generate highly natural and prosodically rich speech. Leveraging a novel two-stage AR-NAR pipeline, MARS5 excels in producing speech for diverse and challenging scenarios.

Key Features

• Advanced Prosody: Generates speech with natural intonation and rhythm.
• Two-Stage Pipeline: Combines autoregressive and non-autoregressive models for high-quality output.
• Minimal Input Requirement: Requires only 5 seconds of audio and a text snippet.
• Customizable Output: Control prosody with punctuation and capitalization.
• Deep Cloning: Enhanced quality with reference transcript for speaker identity.

Use Cases

• Sports Commentary: Generate dynamic and engaging sports commentary.
• Anime Dubbing: Create expressive and character-specific voices for anime.
• Voice Cloning: Clone voices for various applications with high fidelity.
• Interactive Voice Response (IVR): Enhance customer service with natural-sounding automated responses.
• Audiobook Narration: Produce professional-quality audiobook narrations.

Pricing

MARS5 TTS is open-sourced under the GNU AGPL 3.0 license. For commercial inquiries or to license the closed-source version, please contact [email protected].

Teams

CAMB.AI is a globally distributed team of experts, including Interspeech-published researchers and ex-Siri engineers from Carnegie Mellon. We are dedicated to advancing speech synthesis technology and are actively hiring. Interested candidates can reach out to [email protected] for more information.

Join our community on our Forum and Discord to share feedback, suggestions, or questions. Support us on Ko-fi to help us continue our work in making everyone's voice count.