Nvidia announced the launch of a new artificial intelligence model named Nemotron 3.5 Transcribe, which enables converting audio recordings into written text in over 40 languages. This release aims to develop speech recognition and processing applications across various sectors, with the ability to distinguish up to eight speakers in a single recording.
##
In this regard, the tech website TechCrunch stated that the model is available for free with open weights and can process audio recordings up to an hour long. It also provides translation capabilities between English and 31 other languages, expanding its usage scope in multilingual environments.
##
The company added that it unveiled alongside the model launch an updated version of the open-source audio dataset Granary to support developers in building specialized applications. Targeted uses include transcribing business meetings and converting medical conversations into written records, as developers can integrate the model with the dataset and NeMo Speech tools.
##
It was noted that the model comprises 30 billion parameters and can run on a single computer chip. It also supports diverse platforms including the Jetson AGX Thor robotic computing unit, DGX Spark devices, up to Blackwell chips designed for data centers, a step falling within the company's trend of providing models with open weights.
##
The model currently does not support live audio streaming processing, as its operation is limited to saved audio recordings. This feature allows developers to access and customize the trained components of the model to suit the needs of their applications in various technical and industrial fields.
Source: Syrian Arab News Agency (SANA)
Comments (0)