NetEase open-sources a TTS engine with feelings

What it does

EmotiVoice synthesizes speech in English and Chinese with a library of over 2,000 voices. Its standout feature is emotional control: you can prompt it to sound happy, sad, angry, excited, and more, instead of settling for neutral robot narration. The project provides a web interface, scripting tools, and an OpenAI-compatible HTTP API for generating audio.

The interesting bit

The architecture is based on PromptTTS research and decomposes style into pitch, speed, energy, and emotion factors. It also supports voice cloning with personal data, offering recipes for DataBaker and LJSpeech datasets if you want to roll your own speaker.

Key highlights

• 2,000+ built-in voices with prompt-driven emotion and style control
• Voice cloning pipeline with personal data recipes included
• Web UI, Mac desktop app, and OpenAI-compatible HTTP API available
• Docker image with GPU support for deployment
• Apache-2.0 license, though the interactive page has a separate user agreement

Caveats

• Language support is currently limited to English and Chinese; Japanese and Korean are on the roadmap but not yet available
• The style prompt system handles pitch, speed, energy, and emotion, but the README explicitly notes gender is not used as a controllable factor
• Docker deployment requires an NVIDIA GPU

Verdict

Worth exploring if you need expressive, open-source TTS with prompt-level emotional control in English or Chinese. Give it a pass if you require Japanese, Korean, or built-in gender modulation.