Speech system for emotional voice
Turn text into emotional speech, save emotion vectors, keep recent audio in a library, improve a recording, pull audio out of a video, or transcribe speech.
Core Capabilities
Speech generation, duration control, a personal library, emotion vectors, and audio tools.
Emotion Control
8 distinct emotions via vector embeddings or reference audio cloning
Fast, high-quality inference
Speech is generated with hardware acceleration and one stable quality profile
Zero-shot Voice Cloning
Clone any speaker with just 5 seconds of reference audio
Personal library
The last ten speeches with their settings, plus 100 MB for your files
Speech tools
Improve audio, extract a soundtrack from video, and transcribe several languages
Flexible download
WAV, MP3, FLAC, OGG, and M4A at low, medium, high, or original quality
Speech duration
Shorten or lengthen the speech from 0.5× to 2×
Exact pronunciation
Set Chinese Pinyin, English phonemes, or Japanese kana on a word
System Architecture
End-to-end pipeline from text input to emotional speech output
Explore
Everything you need to understand, test, and deploy the system
Generate speech
Choose text, voice, duration, and emotion, then create the audio
Library
The last ten speeches and up to 100 MB of personal files
Speech samples
Audio samples across emotions and evaluation sets
Speech tools
Improve quality, extract audio from video, and transcribe speech
Custom voice
Generate speech from a sample of your own voice
Emotion vectors
Shape a vector, save it, and load the same settings later
Ready to hear emotions?
Try the interactive demo to experience emotional speech synthesis in real-time.
Launch Demo