NVIDIA has launched a new NVIDIA voice AI model designed to make speech transcription and translation easier for developers. Called Nemotron 3.5 Transcribe, it can process audio in 40 languages and identify different speakers.
Released on October 28, 2026, the model also supports speech translation and word-level timestamps. It can process recordings of up to one hour. These features could help developers build tools for meetings, customer service and multilingual communication.
How NVIDIA’s new voice AI model work
Nemotron 3.5 Transcribe converts recorded speech into text. Developers can choose how they want the transcript to appear. For example, they can preserve every spoken word or produce cleaner, more readable text.
The model can also identify different speakers and add timestamps to individual words. This could make it easier to review long interviews, meetings and recorded calls. Developers can also provide relevant terms to improve recognition of specialised vocabulary.
NVIDIA voice AI model supports 40 languages
Multilingual support is one of the model’s key features. Nemotron 3.5 Transcribe supports transcription in 40 languages. It can also translate speech between English and 31 supported languages.
These capabilities could help developers build tools for international teams, media organisations and multilingual customer support. Businesses may also use the model to reduce the need for separate transcription and translation systems.
However, accuracy can vary by language and task. Developers should test the model with their intended users and recordings before deploying it.
Why open-weight AI matters
NVIDIA has released the model weights and supporting resources for developers. This allows teams to download, customise and deploy the model under its applicable licence terms.
Open-weight models offer developers greater control over customisation and deployment. They can also help teams adapt AI tools to specialised vocabulary and specific business needs.
However, running the model requires suitable computing resources and technical expertise. Its availability does not mean every developer can deploy it without additional infrastructure.
What the model could mean for voice AI
Voice AI is becoming increasingly useful in workplace software. Companies already use speech technology for meeting notes, customer support and document preparation.
A model that combines transcription, translation and speaker identification could simplify the development of these applications. It could also help make recorded conversations easier to search and review.
However, Nemotron 3.5 Transcribe processes recorded audio rather than providing real-time streaming transcription. Its practical impact will depend on accuracy, computing costs and developer adoption.
The next step for multilingual AI tools
NVIDIA’s release gives developers another option for building speech-processing applications. Its support for multiple languages and speaker identification could prove useful across several industries.
The bigger test will be how well the model performs in everyday situations. For now, it represents another step towards flexible voice AI tools that developers can adapt to their needs.
Read more : DeepSeek V4.1 Flash update
