Neuphonic

Neuphonic is a company specialising in voice synthesis that uses language models to generate synthetic speech with natural characteristics. It provides voice synthesis services through cloud API and models executable locally on devices. The system includes voice cloning capabilities from brief audio samples.

The platform offers a managed text-to-speech API with support for over 10 languages, including English, Spanish, Arabic, Chinese, Hindi and regional dialects such as Gujarati and Telugu. The cloud infrastructure is deployed on Kubernetes through Google Cloud Platform, designed to handle high concurrency in enterprise implementations with low latency.

Neuphonic features NeuTTS Air, an open-source voice synthesis model with Qwen-based architecture of 0.5B parameters, licensed under Apache 2.0. This compact model runs on CPUs in real-time without requiring GPUs, operating on devices such as phones, laptops and Raspberry Pi-type boards.

NeuTTS Air uses NeuCodec as audio codec. Generated audio files include watermarks for identification. Local processing allows voice data to remain on the user's device without being transmitted to external servers, functioning both online and offline.

Models can be integrated into voice applications, conversational assistants, IoT devices and enterprise systems through APIs with SDKs available in Python and JavaScript.