Vagdhenu

Vagdhenu is a text-to-speech (TTS) model designed for synthesizing Sanskrit chant, specifically referred to as Sanskrit chant (pārāyaṇa) TTS. The model provides single-speaker audio synthesis and is noted for its accurate rendering of conjuncts, including retroflex aspirates, which are stated to render with 100% correctness. According to the evidence, the model achieves a mean opinion score (MOS) of approximately 4.6 as evaluated by expert listeners, indicating a high level of perceived audio quality in its intended use case. The model weights are available for use and are described as suitable for integration with the F5-TTS library. Instructions are referenced for using Vagdhenu with libraries, inference providers, notebooks, and local applications, though specific code snippets are not provided in the evidence. The model is distributed under the Apache 2.0 license, making it available for open-source use and modification. Vagdhenu is intended for applications that require the synthesis of Sanskrit chant, and its features are tailored to accurately reproduce the nuances of Sanskrit pronunciation. The evidence does not specify a particular user group, but the focus on Sanskrit chant synthesis and the mention of expert listener evaluation suggest it may be of interest to those involved in Indic language technologies, research in speech synthesis, or the digital preservation and presentation of Sanskrit texts. The tool is hosted on Hugging Face, where users can access the model weights and documentation for integration into compatible workflows. There is no mention of pricing, so only the open-source Apache 2.0 licensing is confirmed by the evidence.