Earnings Voice Stress Dataset for Investors

---

Every sentence, for every speaker, on every earnings call, for the R3K transcribed.  Repetitions, filler words, hedging words and other aspects of word cadence are tracked and scored.  SCA transcribes the earnings calls itself - we want all that linguistic messiness that are scrubbed from commercial transcripts.

Using a speaker specific baseline, the audio for each sentence is analyzed at the 22 (µs) micro -second level.  Features such as speech rate, vocal tremors, energy and imbalance are generated.  Composite scores are created which map to the speaker's emotional state.

Do stocks of companies whose executive speak with confidence outperform?  How about on topics like operating results, M&A, or earnings?  We provide Jupyter notebooks demonstrating the empirical value of the speaker's emotional state when discussing critical topics.

Each sentence is transcribed and coupled with 19 proprietary voice features generated for every speaker and every sentence on every FOMC press conference since April 2011.

14 proprietary linguistic features generated for every speaker and every sentence on every FOMC press conference since April 2011.

For each speaker on the press conference, the average Voice and Linguistic Feature - useful for creating baselines