Meetings

Record.
No bot in the call.

Your microphone and the system audio are captured on your PC and turned into one transcript with numbered speakers. Nothing joins the call.

Get NexusVox Also: dictate anywhere ›

Ctrl Shift F9

Illustration The meeting hotkey.

to transcribe a 30-minute recording on a CPU: 100 s; Ryzen 7 5800X, 16 threads, no GPU
faster than the recording is long, on a CPU: 18×; Ryzen 7 5800X, 16 threads
of memory at the peak, for that 30-minute file: 2.25 GB; Ryzen 7 5800X, 16 threads

Measured on one PC. Details and sources ›

Everything meetings do

Meeting recording

A recorder, not a bot.

Your microphone and the system audio are captured on your PC. Start and stop in the Meetings tab. Ctrl+Shift+F9 stops a meeting from anywhere. Dictation keeps working meanwhile.

Nobody is added to the call.
While the consent reminder is on, which it is by default, the hotkey opens the Meetings tab so you see the reminder first. You can switch the reminder off in Settings.

Ctrl Shift F9

Illustration Default keys. They stop a meeting from anywhere; starting goes through the Meetings tab while the reminder is on.

Speakers

Who said what.

Your microphone is Speaker 1, or a name you choose. The other participants follow as Speaker 2, Speaker 3 and so on. Up to four voices on the system audio are told apart.

Speaker detection needs the Parakeet GPU container. Without it, everyone else is Speaker 2 and the transcript still completes.

Speaker 1 (you)
Speaker 2
Speaker 3
Speaker 4
Speaker 5

Illustration The labels a transcript uses: you, and up to four other voices. Your own voice can carry a name instead.

Dictation filter

Your dictation stays out.

The microphone also hears what you dictate during a call. By default NexusVox leaves those words out of the meeting transcript.

You dictate during a call
NexusVox notes when
Those words stay out of the transcript

Illustration Default behaviour; switch it off in Settings.

File upload

Got a recording already?

Upload WAV, MP3, FLAC, OGG or WEBM in the dashboard. Everything but WAV needs ffmpeg on your PATH. Switch on “Detect speakers” and the transcript comes back as Speaker 1: … paragraphs, one per turn.

Speaker detection for uploads needs the Parakeet GPU container and tells at most four speakers apart; with more, voices get merged.
Optional text cleanup with the local language model.

to transcribe a 30-minute recording on a CPU: 100 s; Ryzen 7 5800X, 16 threads, no GPU
faster than the recording is long, on a CPU: 18×; Ryzen 7 5800X, 16 threads

Transcribing a recording file on a CPU.

Export

Take the text with you.

Download a transcript as TXT, SRT or VTT. The subtitle formats need Parakeet’s word timing.

Illustration The export formats.

Your data

Yours to keep. Yours to delete.

Meetings play back in the dashboard. Switch off “Store meeting audio” and only the transcript stays. Delete a meeting and its audio with one click.

Play back
Delete with its audio
Keep only the transcript

Illustration What you can do with a stored meeting.

What the numbers mean in practice

Recording file

Example

A 30-minute recording file is transcribed in 100 s on a CPU.

Measured on one single-track file, such as an upload. A meeting recorded in NexusVox has two tracks, your microphone and the system audio, which are transcribed separately, so it takes longer; that case was not measured. Speaker labels need the GPU container.

to transcribe a 30-minute recording on a CPU: 100 s; Ryzen 7 5800X, 16 threads, no GPU
faster than the recording is long, on a CPU: 18×; Ryzen 7 5800X, 16 threads
of memory at the peak, for that 30-minute file: 2.25 GB; Ryzen 7 5800X, 16 threads

30 min Length of the file

→

100 s Time to the transcript

An example from our own measurements, not a customer story.

Fewer threads

Example

With fewer threads, a recording takes longer.

The same desktop CPU pinned to eight and to four threads, on a 5-minute recording file. A slower CPU will be slower still. The real-time factor is processing time divided by audio length: lower is faster.

real-time factor for a 5-minute recording on 8 threads: 0.067; Ryzen 7 5800X, 8 threads
real-time factor for a 5-minute recording on 4 threads: 0.296; Ryzen 7 5800X, 4 threads
of memory with the speech model loaded: 777 MB; Ryzen 7 5800X

0.067 8 threads, real-time factor

→

0.296 4 threads, real-time factor

An example from our own measurements, not a customer story.

Before you record

Tell everyone first. In Germany, recording the spoken word of others without their consent can be a criminal offence (§ 201 StGB), and other countries have comparable rules. This is not legal advice; you are responsible for using the feature lawfully.

Use a headset. With speakers, the microphone also picks up the other participants. Bluetooth headsets lose quality while the microphone is in use: use the PC’s microphone, a wired or a USB headset instead.

Ready to try it?

One download for dictation and meetings. Windows 11, free for private use.