Fonetic, a Spanish company that provides speech analytics and voice recordkeeping software for compliance purposes, is opening its doors in the U.S. today, in the Wall Street area. It's targeting large New York banks first.

The company's software lets banks and trading firms index and catalog calls, chats and emails that must be stored for Dodd-Frank and Mifid compliance, and search for records based on criteria such as banks involved, markets/exchanges used, products (equities, commodities, etc.), traders involved, actual trades, date range, type of activity, and companies involved.

Fonetic's solution can work with 79 languages, understanding dialects, accents and poor pronunciation (e.g., an English trader speaking poor Italian). It also understands slang, turns of phrase and 'in-house' trading terms. For instance, in England the expression, "I'll meet you for a drink" can refer to a bribe, whereas in the U.S. it simply means what it says. The company embeds Nuance and Genesys speech recognition technology in its software to turn the speech into text.

The first use case for this software in the U.S. is trading floors that need to comply with Dodd-Frank and Mifid rules that call for storage and recall of customer conversations. But the technology could also be extended to the back office and call centers, according to Simon Richards, head of global operations at Fonetic.

The software is currently used by BBVA and Santander. Smaller banks would most likely use Fonetic's hosted service.

Digital Reasoning and OP3Nvoice announced yesterday a partnership to offer similar technology.

Fonetic's software takes in feeds of voice calls and emails from a bank's systems as well as any specific jargon the company uses. It makes the content searchable by a compliance director.

Richards concedes that correctly linking emails with related phone calls is not easy, especially where the ticket number is not provided.

"The honest answer is it's impossible," Richards says. "If anybody says 100% that they can do this, that's a lie. We don't say that. We do say that if it exists, we've captured it. If there's no reference to the trade ID, depending on what information the regulator is asking for, that will define the narrowness of the search. Because this is speech and a linguistic model and there's slang and jargon, what you'll get in search where you don't know ticket number is likelihood. From a compliance point of view, we've at least found the needles in the haystack. And the first needle is most likely to be the record you're looking for."

The next use for this software is detection of anger and frustration in customer calls, to identify and start to eliminate problems in the contact center.