Everything you need to know about SpeakSharp — how it works, what it measures, and how to get the most out of your practice sessions.
Getting Started
SpeakSharp is an AI-powered speech coaching tool for anyone who needs to present, pitch, teach, or speak in public. It records your practice runs, analyzes your delivery across 6 metrics, and gives you specific, data-backed coaching — not generic tips.
No. SpeakSharp runs entirely in your browser. Just open the app, allow microphone access, and start recording. No downloads, plugins, or app installs required.
Create a free account on the app page — just an email and password, no credit card needed. You get 5 free analysis sessions. Click the microphone button, speak for at least 8–10 seconds, then tap again to stop. Your results appear in seconds.
An account lets us track your free trial usage fairly and keep your session history. Your presentations and scores are saved so you can track progress over time. We never sell your data or send marketing emails.
Recording & Analysis
Each recording is scored on 6 delivery dimensions:
Pacing — words per minute (ideal: 130–165 WPM)
Filler Control — um, uh, like, basically, so, you know
Tone — pitch variation measured in semitones (ideal: 2–5 st)
Conciseness — average sentence length (ideal: 12–18 words)
Structure — variety of sentence lengths (CV-based)
Clarity — Flesch reading ease (ideal: ≥65)
An overall score (0–100) is calculated from these metrics.
At least 8–10 seconds (roughly 25–30 words) for the system to produce meaningful metrics. For the most useful coaching, aim for 60–90 seconds — long enough to show real patterns in your pacing, filler usage, and sentence variety.
Yes. Below the microphone button there's a drag-and-drop upload zone. You can drop or browse for any common audio format (MP3, WAV, M4A, WebM). The same analysis pipeline runs on uploaded files.
Yes. Click + Add a script to open the script panel. Paste your text, or import a .txt or .pdf file. SpeakSharp will compare your recorded speech against the script and include content-specific feedback in the coaching — deviations, key points covered, and suggestions for tighter alignment.
SpeakSharp records in WebM/Opus format via the browser's MediaRecorder API. This is automatically sent to the analysis server — you don't need to worry about formats.
Scoring & Metrics
The overall score is a weighted average: Pacing (30%), Filler Control (25%), Tone (20%), Conciseness (10%), Structure (8%), Clarity (7%). Each sub-score uses piecewise linear scoring tuned to ideal speech ranges used by professional coaches and communication researchers.
SpeakSharp detects: um, uh, like, basically, so, and you know. Context-aware filtering removes false positives — for example, "I'd like to address" does not flag "like" as a filler. Click any filler word score card to see a timeline of exactly when each filler occurred.
The Structure score balances two things: how many sentences fall in the ideal 12–18 word range (75% of the score) and how varied the lengths are (25%). If most sentences are in the ideal range but all exactly the same length, you'd score around 75. To hit 90+, aim for a mix of short punchy sentences and longer explanatory ones, with most in the 12–18 word band.
When you record multiple attempts in the same session, green ▲ and red ▼ arrows show how each score changed from your previous attempt. This lets you quickly see which areas improved and which regressed. Deltas reset when you create a new presentation or reload the page.
Yes — click any score card (Pacing, Filler, Tone, etc.) to open a drill-down modal with a visualization of that metric and personalized AI coaching advice for it. Clicking the Overall card scrolls to the full coaching block.
Privacy & Account
Audio files are sent to the server for transcription and analysis. A copy is kept on the server temporarily to enable in-session playback. Transcripts and scores are saved locally in your browser (localStorage) and linked to your account for session history — not on remote servers.
Contact us at [email protected] and we'll delete your account and all associated data within 48 hours.
After 5 free sessions, you'll need to upgrade to a paid plan to continue recording new attempts. You can still review all your past attempts and scores. Upgrade options are on the pricing page.
Troubleshooting
This usually means the recording was too short (under 6 meaningful words) or the audio was too quiet. Try speaking for at least 10 seconds and make sure your microphone is not muted or set to low volume. Move closer to your mic if needed.
Click the lock/camera icon in your browser's address bar and set Microphone to "Allow", then reload the page. On mobile, check Settings → Safari/Chrome → Microphone.
Tone is measured as the interquartile range of your pitch in semitones. A score of 2–5 semitones indicates natural conversational expressiveness. Scores below 2 suggest a flat delivery (monotone), while scores above 8 may indicate over-expression that can be hard to follow. Recording in a quiet room also helps — background noise can affect pitch detection.
AI coaching is generated by GPT-4o-mini and may occasionally be too generic or slightly off-target. If this happens, try recording a longer clip (60+ seconds) for more data points. You can also email us at [email protected] with the specific feedback — we use these reports to improve the prompts.