New Phonon

Phonon is a set of small-scale text-to-speech models launched by Gradium; the Hugging Face page provides an online demonstration of these models.

What makes it special is that the entire process takes place entirely within the browser: when you open the application for the first time, the model is downloaded to your local cache. After that, you can simply enter the text, select the language and the tone of the audio, and click to start playing the content in real-time or download it. During the whole process, neither the text nor the sound leaves your device, which makes it particularly suitable for scenarios where privacy is a concern.

The model is available in three different specifications. The smallest version, Phonon 31M, contains only 31 million parameters; after being quantized using an 8-bit format, its size amounts to approximately 65MB upon download. There are also a 40-MB English version and a 40-MB multilingual version. The multilingual version supports five languages: English, French, German, Spanish, and Portuguese; however, Chinese is not currently supported. The model is open-sourced under the Apache 2.0 license, and developers can use the `phonon-tts` package available on npm to integrate it into their own web applications.