Speechmatics Launches Global English, an Accent-agnostic Language Pack for Speech-to-text Transcription

February 22 2018, 03:10
Speech recognition technology specialists Speechmatics announced the launch of Global English, a single English language pack supporting all major English accents for use in speech-to-text transcription. According to the company, Global English (GE) was trained on thousands of hours of spoken data from over 40 countries and tens of billions of words drawn from global sources, making it one of the most comprehensive and accurate accent-agnostic transcription solutions on the market.

Traditionally, speech recognition has dealt with variations in language by producing a different language pack for every distinct accent or region. However, this meant a whole new set of models trained on data from that particular subset of speakers of the languages. With the launch of GE, Speechmatics is aiming to democratize speech-to-text transcription to overcome industry-wide issues where there are multiple English accents in one recording. Thus, providing a far more accurate, consistent and cost-effective solution.

Speechmatics performed a series of tests, comprised of approximately 4 hours of diverse audio and transcribed text. Accented test files included variations in gender, age and region. "When tested against providers of similar solutions, GE consistently produced more accurate transcriptions. Compared directly, GE was between 3% and 55% more accurate than all Google's Cloud Speech API accent-specific language packs and between 5% and 23% more accurate than IBM's Cloud US English language pack," the company states.

"Speech recognition has advanced hugely in recent years, making GE possible. The team has been gathering data from a wide range of sources and taking advantage of the astonishing rise in computer power, allowing them to train bigger models, based on more data, capable of supporting more variations. Speechmatics has now built 72 unique languages, more than any other provider on the market, including Amazon, Google, Nuance, Microsoft and IBM. With the modern neural network architectures capable of generalizing across variations in speech by using representation learning, Speechmatics was able to generate the accuracy of multiple specialized models all in one language pack," Benedikt von Thüngen, CEO at Speechmatics, explains

"At Speechmatics, we have historically produced North American, British and Australian versions of the English language packs, as well as domain-specific language packs. Applications include broadcast, compliance, speech analytics, call recording and meeting transcription among others. While a traditional British language pack does indeed perform better on British accented speech than say, a traditional North American language pack would, there's still tens of distinct British accents to address. And so, we realized we needed to come up with what we like to call 'One Model to Rule Them All' - an accent-agnostic language pack that is just as accurate at transcribing Australian accent as it is with Scottish."

Tom Ash, Speech Recognition Director at Speechmatics, commented: "In the UK alone, there are about 56 main 'accent types', and the concept of having one language pack per accent or region is very outdated. Bearing in mind that we live in an increasingly connected and mobile world, we need our tech to reflect that. We've all heard stories about people being misunderstood by their personal voice assistants or closed captioning getting something awkwardly wrong. While a lot of these stories are humorous, it's ultimately highlighting a big issue. We're hoping that Global English will inspire others to become more flexible and fair when it comes to people's accents."
related items