Question 1

What is a good talk-to-listen ratio on a sales call?

Accepted Answer

The figure most often quoted from call-recording research is around 43 per cent talking for the seller on won deals, with the buyer doing the majority of the talking. Treat that as a direction rather than a target: discovery calls should sit well below half, demos legitimately run higher because someone has to explain the product, and a negotiation is a conversation rather than a presentation. What is consistently bad is a long unbroken monologue, which is why this tool reports your longest one separately from your overall share.

Question 2

Does my audio get uploaded anywhere?

Accepted Answer

No. Recording is captured through the Web Audio API and reduced to loudness numbers in memory as it arrives; a file you pick is read with the browser's own file reader and decoded locally. There is no upload, no transcription, no server call and no account. You can verify it rather than take our word for it: open your browser's developer tools, watch the Network tab, and run an analysis. Nothing carrying audio goes out, because the analysis code has no network access in it at all.

Question 3

Can this tell two speakers apart?

Accepted Answer

Only when the recording genuinely keeps them apart. A single mixed track cannot be reliably split into two speakers by loudness alone, and this tool does not pretend otherwise — with mixed audio it reports speech versus silence for the whole recording, which is exactly right if you record only yourself. If the file has two channels carrying different audio, as call recorders that put each side on its own channel produce, it analyses each channel separately and says so. It also checks whether the two channels are actually different, because plain mono duplicated across a stereo pair would otherwise invent a 50/50 split out of nothing.

Question 4

How does it decide what counts as speech?

Accepted Answer

It splits the audio into 20 millisecond frames, takes the RMS energy of each one, and calls a frame speech when it sits far enough above that recording's own noise floor. The threshold is derived from the material rather than fixed, because an arbitrary file could be recorded at any level. Runs shorter than 140 milliseconds are discarded as clicks and breaths, and pauses shorter than half a second are kept inside the same turn rather than splitting it in two. Every one of those numbers is shown with the results.

Question 5

Why does silence matter so much in this kind of measurement?

Accepted Answer

Because it is the easiest thing to get wrong. If a measurement keeps crediting the last person who spoke while nobody is speaking, a quiet stretch inflates whoever talked most recently. We hit exactly that in the Craqly desktop app: talk ratio kept climbing during silence, and a session that was really about nine per cent one speaker was reporting ninety-six. The fix was an energy gate that stops attribution when nothing is above the floor, and the same gate is the core of this tool.

Question 6

What counts as a monologue?

Accepted Answer

Here, a continuous stretch of speech with no pause longer than half a second. The Craqly desktop app uses a 60-second threshold before it flags a monologue live during a call, which is a reasonable line for a sales conversation: past a minute without a break you are presenting rather than talking with someone. This tool reports your longest stretch as a number and leaves the judgement to you.

Talk ratio analyzer

The audio never leaves this tab, and you can check

Common questions

Related