Voice Intelligence — Dima Perkis

Voice Intelligence

Linguistic analysis of 1,438 recordings — 222 hours of meetings, coaching sessions, and solo notes. mlx-whisper transcription, LLM-powered speaker attribution, hedge rate tracking, acoustic feature extraction, and a RAG search layer across 1.95 million words.

Read the story ↓

What It Is

A quantified self-communication system. Every SuperWhisper dictation, Just Press Record capture, and Voice Memo gets transcribed locally via mlx-whisper, classified by type (meeting, coaching, solo note, interview), enriched with speaker attribution, and analyzed for linguistic markers — hedge rates, filler words, F0 pitch patterns, speaking pace.

What It Found

The “like” acceleration was the headline: usage climbed from 3.8 per 1,000 words to 27.9 over 16 months. Solo recordings showed half the hedge rate of meetings (7.5 vs 13.8), and conversations with my executive coach showed a consistent +39.6 Hz pitch elevation — measurable engagement.

The Stack

Scale

Metric Value
Total recordings 1,438
Audio hours 222
Word count 1.95M
Speakers enrolled 16
ChromaDB chunks 8,688
Date range Dec 2024 — May 2026

Tools that helped

Python Claude