NbAiLab/nb-tts-voxcpm2-voices-2607-demo

Opplest · VoxCPM2

An interactive Norwegian text-to-speech demo for NbAiLab/nb-tts-voxcpm2-voices-2607, a VoxCPM2 checkpoint fine-tuned on two Norwegian Oslo-area voices.

What it does

• —Generates natural Norwegian speech at 48 kHz.
• —Offers a female and a male Oslo voice.
• —Reliably handles medium and long passages: text is split at sentence boundaries, synthesized in sequence, and joined with short natural pauses.
• —Includes one-click examples for quick listening and comparison.

The Space is configured for an NVIDIA A10G GPU and has a runtime idle timeout of one hour, so it automatically sleeps when it is not being used. It uses Nano-vLLM's optimized streaming engine (with a PyTorch fallback) to reduce generation latency. Each result reports its synthesis time, real-time factor (RTF), and speed relative to real time.

Responsible use

This project is intended for research and creative text-to-speech use. Do not use the voices for identity verification, authentication, or deceptive impersonation.

Built with OpenBMB/VoxCPM.