Talk-to-Listen Ratio Calculator and Transcript Grader
Paste a speaker-labeled transcript. Get talk-time share per speaker, longest monologue, question count, and filler-word density scored against the Gong 43/57 sales benchmark. Runs in your browser. Nothing is uploaded or stored.
—
Overall grade
Talk time share
Rep 50%
Prospect 50%
0 rep words
Target: 43% rep
0 prospect words
Longest rep monologue
0sec
Questions asked by rep
0
Filler words (rep)
0(0.0%)
Total speakers detected
0
What to fix on the next call
Want this graded automatically after every call?
Numi runs every recorded sales call through ratio analysis, monologue detection, and discovery-quality scoring. It connects to Fathom, Gong, and Fireflies and flags coaching moments without anyone watching the recording.
Paste a transcript where each line starts with a speaker label and a colon. The parser groups every line by speaker, counts words on each side, and computes talk-time share as a proxy for actual talk time. It then runs four secondary checks: the longest continuous run of rep speech (converted from words to seconds at 150 words per minute), the count of question marks in rep lines, the count of common filler words inside rep speech, and the total number of distinct speakers detected.
The grade compares your rep talk percentage to the benchmark for the call type you selected, then adjusts for monologue length and filler density. A discovery call where the rep talks 44% with no monologue above 90 seconds and almost no fillers gets an A. A demo where the rep talks 80% with a four-minute monologue gets a D regardless of how natural the conversation felt in the moment.
The 43/57 benchmark and where it comes from
Gong analyzed over one million recorded sales calls and found that top-performing reps on discovery calls talk approximately 43% of the time and listen approximately 57%. Most reps invert this ratio. The exact number matters less than the direction: reps who win consistently let the prospect carry more of the conversation than the rep does.
The ratio is not a target in itself. It is a proxy for discovery quality. Reps who ask better questions naturally talk less without thinking about the number. The reason this tool also reports monologue length is that a healthy average ratio can still hide a single five-minute monologue that lost the deal. You can hit 43/57 across a 45-minute call and still have wasted the call.
What the metrics mean
Talk time share. Percentage of total words spoken by the rep versus the prospect. The benchmark depends on call type: 43% on discovery, around 60% on demo, around 50% on closing, around 40% on follow-up.
Longest rep monologue. The longest continuous run of rep speech without a prospect turn, converted from word count to seconds. Top reps stay under 76 seconds. The 90-second mark is the practical warning line. Above 180 seconds the rep is almost certainly in presentation mode.
Questions asked by rep. Count of question marks in rep lines. On a 30-minute discovery call, top reps tend to land 11 to 14 open questions. Below 6 questions on a discovery call is usually a presentation in disguise.
Filler words. Count of high-frequency fillers (um, uh, like, you know, basically, literally, actually, kind of, sort of, I mean, right, okay, so) in rep lines. Density above 2% of total rep words is usually noticeable to a listener.
Transcript formats that work
Any line-based format with a speaker label and a colon will parse. Exports from Gong, Chorus, Fireflies, Otter, Zoom, and Google Meet all work out of the box. Timestamps in front of speaker names are stripped automatically, so both of these parse identically:
Rep: Tell me about the current setup.
[00:01:23] Rep: Tell me about the current setup.
If your transcript has more than two speakers, the parser picks the two most-active speakers and treats the rest as background. If you want to force which speaker is treated as the rep, type that exact label into the "Rep label in transcript" field above.
Frequently asked questions
Gong's analysis of over one million recorded sales calls puts the benchmark for top-performing reps at approximately 43% talk time and 57% listen time on discovery calls. The number shifts with call type: demos can run higher rep talk because the rep is walking through product, and closing calls tend to sit near 50/50. The ratio is a proxy for discovery quality, not a target in itself. Reps who ask better questions naturally talk less without thinking about the number.
The tool runs entirely in your browser. It splits the transcript on speaker labels (lines like "Rep:" or "[00:01:23] Prospect:"), groups everything one speaker says, and uses word count per speaker as a proxy for talk time. Word count tracks talk time closely when the speaking rate is similar on both sides, which is the assumption every transcript-only ratio tool makes. Nothing is uploaded or stored.
Any format where each line starts with a speaker label followed by a colon. Examples that work: "Rep: Hi, thanks for taking the time." or "[00:01:23] Sarah: We tried that last quarter." Gong, Fireflies, Chorus, Otter, Zoom, and Google Meet exports all use a compatible structure out of the box. Timestamps in front of the speaker name are stripped automatically. If your file uses a different convention, do a quick find-and-replace on the speaker names before pasting.
A rep can hit a 43/57 ratio on average and still lose deals because the rep's talk time is concentrated in long uninterrupted blocks. Gong's data shows top reps keep continuous talk segments under 76 seconds. A practical rule is 90 seconds: any stretch you spoke for 90 seconds or more without a question is a high-risk zone. This tool flags your longest monologue specifically so you can see whether your ratio is healthy or whether you have one or two presentation-mode segments dragging the call off the rails.
The calculator counts a curated list of high-frequency fillers: um, uh, like, you know, basically, literally, actually, kind of, sort of, I mean, right, okay, and so. Matching is case-insensitive and word-bounded so "actually" counts as a filler but "actual numbers" does not. Filler density is reported as a percentage of total words for the rep. Anything above 2% is usually noticeable to a listener.
Yes, free with no signup, no rate limit, no data retention. The parser runs in your browser using plain JavaScript. The transcript never leaves your device. There is no backend call. Refresh the page and the transcript is gone.
Related reading and tools
Go deeper on talk-to-listen ratios, or analyze your other outbound content: