Speech Craft Analytics provides voice‑tone analytics for earnings‑call transcripts, extracting the underlying conviction, uncertainty, and psychological state behind executives’ words. By applying specialized NLP that focuses on vocal cues rather than just text, the platform gives investment and IR teams a measurable edge in detecting confidence or hesitation that standard transcripts miss. The service is tailored for fundamental investors seeking deeper insight into management sentiment during key financial disclosures.
Funding
Funding not disclosed
Founders
Product
Problem
Investors rely on earnings‑call transcripts, but executives can control the wording while their vocal delivery reveals genuine confidence, uncertainty, or stress that text‑only analysis misses. This creates a blind spot for fundamental research and risk assessment.
Solution
Speech Craft Analytics (SCA) extracts quantitative acoustic, linguistic, and behavioral signals from CEO/CFO earnings‑call audio. By applying speech‑science techniques, machine‑learning models, and proprietary linguistic analyses, SCA converts raw voice recordings into interpretable features such as pitch dynamics, jitter, speech rate, voice arousal, confidence probabilities, and sentiment markers. The platform delivers these metrics at the sentence level, along with event‑study‑ready excess‑return data and decile rankings, enabling investors to incorporate management tone into factor models and trading signals. SCA’s dataset is structured for direct integration with quantitative research pipelines, LLM‑based analysis tools, and risk‑management frameworks, providing an additional layer of insight beyond traditional NLP.
Target Audience
Primary users are fundamental investors, equity analysts, and quantitative research teams that build trading signals, as well as investor‑relations professionals seeking data‑driven assessments of executive communication.
Features
- Acoustic factors including normalized pitch, jitter, shimmer, energy, speech rate, pause structure, MFCC embeddings, and Voice Delivery Quality (VDQ)
- Linguistic factors such as sentiment polarity, hedging, filler words, honesty phrases, complexity indices, and pronoun usage
- Composite behavioral metrics that blend audio and text (e.g., assertiveness, nervousness, arousal, valence, confidence probabilities)
- Sentence‑level dataset with timestamps, speaker IDs, and >200 acoustic and linguistic metrics per utterance
- Event‑level return data with excess returns for multiple horizons, decile assignments, and controls for earnings surprise and reporter effects
- Model‑ready feature format compatible with factor models, stock‑selection algorithms, and LLM/RAG systems
- Visualization and export tools for analysts to pinpoint topics where tone diverges from content