Your browser is out-of-date!

Update your browser to view this website correctly. Update my browser now


Sage Tells FCC About TTS Quality Impediments

EAS manufacturer suggests commission consider Canadian approach

A rack of Sage EAS hardware
FCC officials have been meeting by phone with EAS manufacturers to talk about alerting matters, including multilingual alerts and issues with text-to-speech quality.

Sage Alerting Systems is one of them; the company makes many of the CAP/EAS devices used by U.S. radio stations. Harold Price and Christopher Vournazos spoke with the FCC’s Greg Cooke, James Wiley, Jane Kelly, Austin Randazzo, Steven Carpenter and Jessica Krentz at the staff’s request, according to a summary of the call filed by Sage with the commission.

Two major topics of that talk were multilingual EAS and speech-to-text.

Sage, which is active in the FCC’s Communications Security, Reliability and Interoperability Council, told them it supports the advisory group’s recommendations that the commission refrain from creating additional regulations for multilingual alerting right now, and that use of non-English or multilingual alerting should remain voluntary for EAS originators and participants (such as radio stations).

“The bureau asked about how EAS participants could make use of multiple languages in CAP messages,” Sage continued in the summary. “When alternate language text and audio are provided, the task is easy. Sage plays the provided audio and presents the provided text. Display of the text is a ‘downstream’ issue. CAP allows for the use of an expanded character set (UTF-8), proper display of the text depends on the ability of the downstream device to render the characters using a suitable font. Languages that are read right to left, or vertically, present an additional challenge, but the basic mechanism for transporting the characters is the same.”

When text for a language is provided but audio is not, Sage continued, “it is left to the device to generate audio with text to speech (TTS). TTS is a permitted option in the current Part 11 rules, Sage supplies English TTS with all of its CAP/EAS equipment. Sage supports Spanish as an installable option.” The final case is where an EAS participant wants to use an alternate language, but the originator does not provide audio or text in that language. “The CSRIC report discusses the current state of translation tools, automated or otherwise. Such systems are not in widespread use in the CAP/EAS context at this time,” the company wrote.

And about text-to-speech, Price and Vournazos told the FCC staffers that while the state of TTS is acceptable for some CAP/EAS messages, local place name pronunciation and the use of punctuation and abbreviations are both impediments to quality.

“Originators of CAP messages intended for broadcast via the EAS system are typically not FCC licensees and are not under its control,” the company noted. “The lack of a ‘style guide’ for message originators, along with a common set of abbreviations and formats for such items as time of day, can result in TTS that is very hard to understand.” The company wondered who is responsible for determining and publishing local and regional place name pronunciations; who could take that data and reformat it as needed by the CAP/EAS vendors; and who could distribute it to thousands of EAS participants and keep it up to date. “While some local areas are beginning to consider such issues, most are not.”

Sage recommended that the bureau, with FEMA, consider a solution used by Canada to address the TTS quality problem.“As in the U.S, CAP messages are relayed to the equivalent of EAS participants via a common server called NAAD, and run by Pelmorex,” it stated. “Many of the CAP messages that are aired in Canada come from its weather service, which provides only text. As in the U.S., the text was in a format with odd punctuation, some abbreviations and many local place names. In the first major use of broadcast CAP messages, in 2015, the overall result using TTS was viewed as unacceptable.”

Sage said a partnership between Environment Canada and Pelmorex defined a new “broadcast text” field, with agreement on formatting and abbreviations. Text is provided by the originator in English, French or both. This helps avoid impact on existing non-broadcast users of the CAP data. Each province was able to provide a set of pronunciations for place names. On receipt of the CAP message, Pelmorex provides a centralized TTS conversion and builds an audio file; the CAP device at each broadcaster can then play the provided audio. Any message originator can use the same broadcast text field and have the TTS provided by the common TTS server.

Thus, Sage concluded, while the problem of formats, abbreviations and pronunciations is not eliminated, “it is bounded and centralized, and the individual stations have no responsibility for maintaining TTS lexicons. Any improvement in the centralized TTS server results in immediate improvement at all stations. Overall system cost is reduced, and overall system quality is improved. The province of Alberta has always used a similar server-side TTS system in its provincial CAP server.”

A similar ex parte summary was filed by manufacturer Trilithic, read it here, and the Sage filing is here.