Archive
The whole site in one list, ordered by the date printed on each page.
One page here belongs to 2017. It explains speech-to-text engines trained on trading audio as they stood that year, and its figures and regulatory timing are those of the months before MiFID II took effect. The pages written in 2026 describe voice desks, speech recognition and recordkeeping as they stand now, including a look at retail trading in Spain and what a taped broker call protects.
Rules on recording and retention change, and so do recognition error rates. A date on a page marks the point at which it was accurate. The same pages grouped by subject sit in Speech technology, Voice in the markets and Records and rules.
Newest first
-
Accents, crosstalk and the phone line: where speech recognition slips
Why speech recognizers make more errors on some voices and on trading-desk audio: training data, narrowband phone lines, overlapping speech and what firms do about it.
-
How long trade recordings must be kept
Five years under MiFID II, one year for oral records at the CFTC, three for FINRA taping firms: retention periods for trading calls and when the clock starts.
-
How voice trading desks still work
Turrets, private wires, the squawk box and the voice broker: how trades are still agreed by phone, and why voice survives in large and unusual deals.
-
Trade call recording rules: MiFID II, the CFTC and FINRA
What MiFID II Article 16(7), CFTC Rule 1.35 and FINRA Rule 3170 each require firms to record, who they cover, and where the European and U.S. rules part ways.
-
Voice on the record, and never borrow to win it back
Why European brokers must record client calls, what Spain's CNMV says about CFDs and finfluencers, and why borrowed money never belongs in a trading account.
-
What transcription changed for trade surveillance
Firms recorded every trading call for years and listened to almost none. Automatic transcription put voice into the same review as chat, with new limits.
-
Word error rate, and what it leaves out
How word error rate is calculated, why it can pass 100 percent, how normalization moves it, and why a trading transcript needs more than one number.
-
Speech-to-text built for the trading floor
Why general speech recognition stumbles on trader talk, what an engine trained on market audio does differently, and why MiFID II makes searchable calls matter.
