Nvidia has introduced Nemotron 3.5 Transcribe, an open speech recognition model capable of transcription, translation, and speaker recognition. According to The Decoder, the model can also take instructions via text prompts, enhancing its versatility for various language tasks.
The Decoder further reports that Nemotron 3.5 leads multiple benchmarks while using only one sixth of the active parameters compared to its strongest open-source competitor. Nvidia is also providing a large dataset exceeding 800,000 hours of speech, along with tools for fine-tuning the model to specific needs.
This development is particularly relevant for Japanese markets where efficient, multilingual speech recognition technologies are in increasing demand across sectors such as FX trading, crypto communication, and equity research, enabling faster and more accurate data processing.