Backed by Wharton, Stanford & TED research

Quantify your executive presence. Then improve it.

Lucid is a Mac desktop application that uses research from the world's leading experts to passively monitor your executive presence while on calls, and gives you expert, personalized insights to improve your executive presence.

5 dimensions analyzed
0 bytes to the cloud
100% local processing
Download for Mac Free  ·  macOS 12+
Lucid Coach EP 82
82
EP
Confident
Strong delivery. Watch hedging: replace "I think" with direct assertions to push toward Distinguished.
Filler Rate
80
Pace
88
Vocal Variety
72
Conciseness
82
Vocabulary
76
Research Foundation

Built on peer-reviewed speech science.

Every dimension Lucid Coach measures is grounded in published research linking specific speech behaviors to perceived authority, competence, and leadership effectiveness.

Wharton School
Filler words reduce perceived competence by up to 20%

Wharton researchers found that filler word frequency is directly and significantly correlated with how competent, credible, and confident speakers are judged to be, independent of the actual content of what they say.

Read the research
Stanford / Executive Communication
Vocal variety is the strongest predictor of speaker ratings

Research on executive communication consistently identifies pitch variation, volume dynamics, and strategic pausing as the primary factors distinguishing highly-rated speakers from monotone ones, regardless of content quality.

Read the research
Published TED Talk Analysis
Top-rated speakers use 40% fewer hedge words than average

Analysis of high-rated vs. average-rated TED Talks found that hedging language ("I think," "maybe," "possibly") was a strong negative predictor of audience reception, while assertive language patterns correlated with higher ratings and virality.

Read the analysis
Lucid Coach translates these findings into real-time feedback. Each dimension maps directly to a research-validated behavior. Each coaching insight maps to a specific, peer-reviewed mechanism for improvement.
The research identifies what matters. Lucid Coach operationalizes it into five precisely-defined, continuously-measured dimensions.
What Gets Measured

Five dimensions. Precisely defined.

Each dimension has a specific operational definition, a measurable target range, and a weighted contribution to your Executive Presence score.

Dimension
What is measured and why
EP Weight
Filler Rate Tracks 12+ filler types Target: under 2/min
Context-sensitive detection of um, uh, like, you know, so, basically, right, actually, and more. "Like" as a comparison ("looks like the data") is not flagged. "Like" as a filler is. Word-level timestamps from Whisper transcription make each flag precise to within the sentence.
Research basis: Wharton study linking filler frequency to perceived competence.
25% of EP score
Pace Words per minute Optimal: 130-160 WPM
Computed from word-level timestamps in every transcript. Fast speech (above 200 WPM) is penalized 2x harder than slow speech because rushed delivery signals anxiety, not expertise. Strategic pauses are tracked and rewarded separately from hesitation pauses.
Research basis: Executive coaching studies linking deliberate pacing to authority perception.
20% of EP score
Vocal Variety Pitch + volume dynamics Detects monotone, uptalk
Parselmouth (Praat wrapper) extracts fundamental frequency (F0) frame by frame. Pitch standard deviation below 15 Hz flags monotone delivery. Rising pitch at clause ends (more than 20 Hz) flags uptalk. Volume dynamics (dB/frame change) measures energy variation.
Research basis: Strongest single predictor of speaker ratings across multiple executive communication studies.
20% of EP score
Conciseness Words per sentence Target: 12-18 words/sentence
Sentences are segmented using pause-based analysis of word timestamps (not punctuation), which captures natural spoken sentence boundaries rather than written ones. Verbosity undermines authority by diluting key points in unnecessary filler words.
Research basis: Executive presence studies linking concise speech to decisiveness perception.
20% of EP score
Vocabulary Power words, hedges, qualifiers MTLD diversity metric
spaCy NLP detects 50+ power words ("will," "commit," "deliver," "ensure"), 25+ hedge phrases ("I think," "maybe," "hopefully"), and weak qualifiers ("just," "a little," "somewhat"). MTLD (Measure of Textual Lexical Diversity) quantifies vocabulary range independent of speech length.
Research basis: TED Talk analysis linking hedge-to-power word ratio to audience reception.
15% of EP score
Precise definitions require precise measurement. Here is every stage of the analysis pipeline, running locally on your Mac.
Technical Architecture

How Lucid Coach works, precisely.

Seven processing stages, all running locally on your Mac. No audio leaves your device at any stage.

01 Audio Capture sounddevice captures 16kHz mono audio from your selected microphone. Raw audio never touches disk.
02 Voice Activity Detection Silero VAD (PyTorch model) identifies speech segments in real time, discarding silence and background noise before it enters the pipeline.
03 Speaker Verification SpeechBrain ECAPA-TDNN verifies 1.2-second audio segments against your enrolled voice profile using 192-dimensional cosine similarity. Only your speech passes through.
04 Transcription faster-whisper (small.en, int8 quantized) transcribes verified speech with word-level timestamps. All transcription runs on-device.
05 Multi-Analyzer Pass Three analyzers run in parallel: FillerDetector (context-aware NLP), LanguageAnalyzer (spaCy for hedges, power words, passive voice), ProsodyAnalyzer (Parselmouth for pitch, volume, uptalk).
06 Score Computation Each dimension is scored 0-100. EMA smoothing (alpha=0.7) prevents score volatility from single utterances. Weighted composite yields the Executive Presence score.
07 Save to Local Database Scores, coaching tip, and anonymized metrics are persisted to a local SQLite file in WAL mode. No transcript text is stored. No audio is stored.
Speaker gate prevents crosstalk
Sandwich recovery, momentum scoring, and anti-momentum logic ensure borderline cases are handled correctly. Your colleague's voice never enters analysis.
Sub-60-second analysis cycle
Each analysis covers approximately 30-90 seconds of your speech, then updates your score. The cycle repeats continuously while you are speaking.
Graceful degradation
Microphone disconnect triggers exponential backoff reconnection. Model load timeouts default to graceful fallbacks. Watchdog prevents analysis hangs.
Scoring System

Four tiers of executive presence.

The EP score ranges 0-100, computed as a weighted average of all five dimensions. Each tier has a specific behavioral description, not just a label.

1 - 54
Building
Active coaching phase. Multiple high-priority patterns identified. Clear improvement path.
Filler rate typically above 8/100 words
55 - 69
Developing
Solid foundation with 1-2 specific patterns to address. Usually pace or vocal variety.
Filler rate typically 3-7/100 words
70 - 84
Confident
Strong communicator. Minor refinements available. Board-meeting ready most days.
Filler rate typically 1-3/100 words
85 - 100
Distinguished
CEO-level delivery. Minimal fillers, commanding pace, precise vocabulary, strong vocal variety.
Filler rate typically under 1/100 words

Score uses exponential moving average (alpha=0.7) to smooth variance between analysis cycles. A single bad segment will not crash your score.

Privacy Architecture

Zero-trust privacy by design.

Privacy is not a policy statement here. It is the architecture. These claims are verifiable.

All processing is local
Transcription, speaker verification, NLP analysis, and scoring run entirely on your Mac via on-device models. No API calls for analysis. No server connection required.
Audio is analyzed and discarded
Audio is processed in memory and immediately freed. The SQLite database contains scores, metrics, and coaching tips only. No audio files. No transcript text in storage.
No account. No telemetry. No analytics.
Lucid Coach has no user registration, no usage tracking, no crash reporting to a remote server, no analytics SDK, and no third-party dependencies that communicate externally.
Verify it yourself
1 Open Lucid Coach. Let it run through one analysis cycle. Note your score.
2 Turn off WiFi in your Mac's System Settings.
3 Speak for another 60 seconds. Watch your score update.
If the score updates while offline, it is not calling a server. That is the only privacy guarantee that matters.
Competitive Context

Lucid Coach vs. the alternatives.

Every executive in a meeting has speech coaching options. Most are expensive, manual, or miss 90% of your speaking time.

Feature
Lucid Coach
Executive coach
Recording yourself
Coverage
Every meeting, every day
1 hour/week at most
Manual, when remembered
Feedback timing
Real time, mid-conversation
Days or weeks later
After the fact, if reviewed
Objectivity
Measured, not subjective
Subjective assessment
Requires self-assessment
Privacy
100% local, zero cloud
Shares confidential conversations
Audio files on device or cloud
Cost
Free
$300-$600/hour
Free but low ROI
Setup
60-second voice enrollment
Months to find and vet
None needed
Ready
Your next meeting starts soon.
Your score should already be running.

It takes 60 seconds to enroll your voice. After that, Lucid Coach runs passively in the background during every conversation. You speak. It measures. You improve.

Download Lucid Coach, Free
macOS 12+  ·  No account  ·  No cloud
Questions

Common questions.

How accurate is the speaker verification?
SpeechBrain's ECAPA-TDNN model is trained on VoxCeleb (1M+ utterances, 7,000+ speakers) and uses 192-dimensional speaker embeddings with cosine similarity verification. Enrollment includes codec-augmented variants (Opus 32kbps and 16kbps) to handle VoIP degradation in Zoom and Teams calls. The system also uses sandwich recovery (retroactively accepting borderline rejections flanked by verified segments) and adaptive voiceprint updating for voice drift over time.
What happens if I have a cold or my voice changes?
The adaptive voiceprint system slowly updates your centroid using high-confidence verifications (75% adapted, 25% enrollment anchor). After 24 hours of no verification (e.g., after a weekend), it blends 50/50 with your enrollment baseline. For significant voice changes, the Settings screen offers a one-click re-enrollment. The morning profile enhancement endpoint also runs daily to refresh the voiceprint.
How much memory does it use?
The ML stack (VAD, Whisper small.en, SpeechBrain, spaCy) loads approximately 1-2 GB of RAM on first use. Lucid Coach monitors system memory and pauses analysis if total system usage exceeds 2.5 GB to avoid affecting your other apps. All models load lazily in background threads. The UI is responsive before models are ready.
Does it work during Zoom/Teams calls?
Yes. Lucid Coach captures audio from your selected input device (typically your microphone), not the system audio output. During Zoom/Teams calls, your microphone audio is the signal: your voice is captured, your colleagues' voices through the speaker are not. Speaker verification further ensures only your voice is analyzed. Voice enrollment includes codec-augmented training samples specifically to handle Opus 16kbps VoIP compression.
Is this actually free?
Yes. All analysis, scoring, coaching, and trend tracking is free. The only optional premium feature is AI-powered daily coaching tips via DeepSeek V3. This requires your own DeepSeek API key (available at a cost from DeepSeek's platform). The template-based coaching tips, which are also included, require no API key.
Getting Started

Running in three steps.

Lucid Coach requires no ongoing configuration. Once you enroll your voice, it works in the background of every conversation automatically.

01
Download and open
Install on macOS 12 or later. No account creation. No email address. The app opens immediately with a setup screen.
02
Enroll your voice (60 seconds)
Read a short passage aloud. The ECAPA-TDNN model creates a 192-dimensional voiceprint. This is the only setup step. It never happens again.
03
Speak. Your score updates automatically.
Open Lucid Coach in the background before your next meeting. Speak naturally. Your EP score updates every 30 to 90 seconds. Close the app when done. No data leaves your machine.
Early Feedback

What users measure in the first month.

Early users share what changed after Lucid Coach gave them a number to work toward. Average EP score improvement in the first 30 days: +16 points.

EP 58 to 79 over 6 weeks
I knew I said "um" a lot. I did not know it was 22 times in a 15-minute standup. Seeing the exact count changed my behavior in a way that two years of vague awareness never did.
VP of Engineering, Series C SaaS company
EP 63 to 85 over 8 weeks
The filler rate is obvious. The vocal variety score surprised me. I sounded confident to myself, but the pitch data showed flat delivery. My board presentations are measurably different now.
Founder and CEO, venture-backed startup
EP 71 to 88 over 5 weeks
The privacy angle was the deciding factor for me. I lead sensitive negotiations. The offline verification test convinced me before I installed it. Score went from 71 to 88 in a month.
Managing Director, investment management firm
Download

Measure what matters.

Download Lucid Coach for Mac. Free, local, and ready to run in the background from your next meeting.

Download for Mac (Free)
macOS 12 or later No account required No data collected