{"id":2012,"date":"2026-10-06T10:30:05","date_gmt":"2026-10-06T10:30:05","guid":{"rendered":"https:\/\/blog.openzeka.com\/en\/?p=2012"},"modified":"2026-10-06T11:06:18","modified_gmt":"2026-10-06T11:06:18","slug":"speech-studio-transcribing-recordings-on-a-jetson","status":"publish","type":"post","link":"https:\/\/blog.openzeka.com\/en\/speech-studio-transcribing-recordings-on-a-jetson\/","title":{"rendered":"Who Spoke, and When? Speaker Diarization on Jetson with NVIDIA Nemotron-3-Diarization"},"content":{"rendered":"

Finding one particular exchange in a meeting, interview or podcast can mean listening through the entire recording. The answer to \u201cwho said what?\u201d is in the file, but reaching it takes time. Cloud services make this easier, but they require you to upload the recording to another server.<\/p>\n

Speech Studio<\/strong> is an open-source tool that does this work locally on an NVIDIA Jetson Orin Nano Developer Kit<\/strong>. It transcribes recordings, separates speakers, summarizes the content and answers questions about the recording. The audio file stays on the device throughout the process.<\/p>\n

At the heart of the app is NVIDIA\u2019s speaker diarization model, Nemotron-3-Diarization<\/a><\/strong>. It answers the question \u201cwho spoke when?\u201d, and it gets its own section below.<\/p>\n

The source code is available on GitHub: github.com\/openzeka\/speech-studio<\/a><\/strong><\/p>\n

The measurements and screenshots in this post were taken on an 8 GB NVIDIA Jetson Orin Nano Developer Kit<\/strong>.<\/p>\n<\/div>

\"Speech<\/span><\/div>

<\/h2>\n

A real recording<\/h2>\n

The main test used a 20 min 30 s recording taken from a conversation between NVIDIA and OpenAI about AI infrastructure<\/a>.<\/p>\n

Speech Studio found 3 speakers and 187 speaker turns<\/strong>. It matched each part of the transcript with a speaker and a timestamp:<\/p>\n

\n

[00:09]<\/strong> Speaker 3:<\/em> So one of the things we wanted to talk about today is the concept of building AI infrastructure at scale.<\/p>\n<\/blockquote>\n

Labels such as \u201cSpeaker 3\u201d can be replaced with a person\u2019s name. The app also produces a summary, key points and action items. When you ask a question about the recording, it points to the relevant passages with timestamps.<\/p>\n

Transcripts and summaries are generated by AI and can contain mistakes. Proper names are especially prone to errors. Timestamps make it easier to check important details against the original recording.<\/p>\n<\/div>

\"Transcript,<\/span><\/div>

<\/h2>\n

The model that separates speakers: Nemotron-3-Diarization<\/span><\/h2>\n

Three different questions you can ask a recording<\/h3>\n

It helps to sort speech AI models by the question they answer:<\/p>\n