Text-to-speech produces voices that narrate e-books and documents, converse with humans as smart assistants or digital avatars, and are part of nearly all digital devices, including smartphones, tablets, and laptops. NVIDIA Riva text-to-speech (TTS) provides human-like synthetic voices based on state-of-the-art spectrogram generation and vocoder models. TTS pipelines are customizable and GPU-optimized to run efficiently in real time.

• SOTA models for generating expressive, human-like voices
• Two out-of-the-box professional female and male voices for US English
• Easy voice and accent fine-tuning with pitch, volume, and duration control for expressivity
• 12X higher inference performance versus existing technologies

Riva embedded delivers real-time, reliable, and best-in-class accurate transcripts and expressive, human-like voices for conversational applications on devices such as delivery robots, intelligent touchless kiosks, vending machines, and virtual assistants for factory, shopping, medical, and smart home devices.

• SOTA, out-of-the-box ASR accuracy with full off-device customization for English, Spanish, Mandarin, Hindi, Russian, German, and French
• Expressive OOTB professional female and male English voices deployable immediately on the device and with the ability to create new brand voices
• Easy integration and reliable, real-time workstation performance in compact, on-device compute and memory
• High privacy with speech data processing on the device
• Deployable on NVIDIA Jetson AGX Xavier, Jetson Xavier NX, Jetson AGX Orin and Jetson Orin NX