Home › Glossary
Glossary
Glossary
Speech software carries a lot of jargon, and some of it means different things to different vendors. Here is exactly what each word means when we use it.
Words used on this site in a specific or unusual way. Every one of them is linked from the first place it appears on a page.
- On-device
- The work happens on the computer in front of you, not on a company’s server. For Lesskeys that means the recording, the speech model and the finished text all stay on your Mac. It is a statement about where the processing happens, which is why it can be checked: turn the network off and it still works.
- Speaker diarization
- Working out who spoke, and when, in a recording with more than one voice. It is a separate job from transcription: transcription asks what was said, diarization asks who said it. The two are combined to produce a transcript with names against each line.
- Dictation
- Speaking instead of typing, so the words appear in whatever app you are already using. It is live and short. Contrast it with transcription, which turns an existing recording into text after the fact.
- Transcription
- Turning a recording that already exists — a call, an interview, a video — into written text. Unlike dictation it happens after the event, and it usually involves more than one speaker.
- System audio
- The sound your computer is playing, as opposed to the sound your microphone hears. Recording system audio plus the microphone captures both sides of a call without anything joining the meeting.
- ASR (automatic speech recognition)
- The general name for software that converts speech into text. Whisper is one such system. “Speech to text” means the same thing in ordinary use.
- WER (word error rate)
- How accuracy is measured. It counts the words a system got wrong — substituted, deleted or inserted — as a percentage of the words actually spoken. Lower is better, and a WER of 2.5% means roughly one word in forty is wrong.
- Whisper
- A family of open speech-recognition models that can run on ordinary hardware, including a Mac. It comes in several sizes; bigger is not automatically better, which is what our own benchmark measured.
- Hotkey
- A key combination that works everywhere, not just inside one app. Lesskeys uses one to start and stop dictation, so you never have to switch windows to talk.
- Voiceprint
- A stored summary of what a particular voice sounds like, used to recognise the same person across different recordings. Lesskeys keeps voiceprints on your Mac and nowhere else.
Free to download. One-time unlock. No subscription. · Pro $49.99 one-time · macOS 13+
FAQ
What does on-device actually mean?
That the processing happens on your own computer rather than on a company's server. The practical test is simple: turn off the network. If it still works, it is on-device.
Is dictation the same as transcription?
No. Dictation is live — you speak and the words appear where you are typing. Transcription turns a recording that already exists into text, usually with more than one speaker in it.
What is a good word error rate?
It depends on the audio, which is why a single number is misleading. On clean speech the best current models sit around 2–3%, meaning roughly one word in forty is wrong. Difficult audio — crosstalk, strong accents, a bad room — is much harder for every system.