XNews

Soprano: 80M-Parameter Lightweight TTS Model with 20x Real-Time CPU Synthesis

Soprano: 80M-Parameter Lightweight TTS Model with 20x Real-Time CPU Synthesis

Soprano is introduced as an ultra-realistic text-to-speech model with only 80 million parameters and under 1 GB memory usage. On an ordinary CPU, it achieves synthesis speeds of 20x real-time and an initial latency below 250 milliseconds. Long input texts are automatically split sentence by sentence for unlimited-length synthesis, and output is delivered at 32 kHz audio quality. The tool exposes an OpenAI-compatible API and also offers a command-line interface and a web UI. It runs across Windows, Linux, and macOS, providing a local, low-resource alternative to cloud-based TTS services.

Source posts