This November, we're thrilled to bring you our latest updates and innovations. It’s been an exciting time for us, particularly in our language packs.

Our aim is to understand every voice, making speech-to-text more accessible and breaking down the barriers created by language. In line with this effort, we've added 14 new languages. Now, with 50 languages we cover over half the world’s population.

Elsewhere, we've seen accuracy levels jump on 20 of our existing language packs, added new information to the JSON output to help customers generate text transcripts based on language properties, and we’ve launched our Real-Time SaaS.

Speechmatics Goes Even More Global: 14 New Languages

In our pursuit to understand every voice, we need to understand more languages. Until recently, our speech-to-text was available to use in 35 languages. Now we’ve taken a huge leap of 14 languages to 50, with an eye to adding even more in the future. Here are the additions: Bashkir, Basque, Belarusian, Esperanto, Estonian, Galician, Interlingua, Marathi, Mongolian, Tamil, Thai, Uyghur, Vietnamese, and Welsh.

With the inclusion of lesser-spoken languages such as Welsh (883,300) and Basque (900,000), we’re working to preserve these cultures with our technology. Right now, this means Speechmatics’ speech-to-text covers 50% of the global population. While it’s our biggest addition ever, we’re aiming to reach 70% usability in the next three years.

20 Languages Receive Improvements

While increasing the number of languages makes our technology more accessible, a focus on accuracy is essential as a differentiator in a competitive industry. To that end, we’ve updated 20 languages, as follows: Latvian, Swedish, Hungarian, Portuguese, Polish, Mandarin Chinese, Arabic, Dutch, Slovak, Bulgarian, Romanian, Slovenian, Lithuanian, Croatian, Malay, Catalan, Czech, Danish, Greek, and Turkish.

Our team’s efforts produced some huge upticks in accuracy including a 6.9% relative decrease in word error rate (WER) for Latvian, 6.6% for Swedish, 6.2% for Portuguese, and the same for Hungarian.

We’ve also seen improved formatting of numeric entities such as dates, currencies, and large numbers for Swedish, Norwegian, and Dutch.

New Language Formatting

Another new update for the quarter sees updates to our JSON output – now at version 2.8. We’re offering more detailed information about properties of the language being used, such as writing direction and word delimiter.

Different properties of language – such as with Arabic text (written right-to-left), or Chinese Mandarin, which has no spaces between words – have historically proven a challenge for speech-to-text. However, with this latest update, our engine will now expose certain information about the language pack in our JSON output. This helps us understand the nuances that come with different dialects around the world.

At the recent IBC show in Amsterdam, we spoke to a number of customers about formatting issues. These marked improvements with delimiters will help them in a variety of areas including compliance, the manual process of checking spaces, and control over punctuation.

The Release of our Real-Time SaaS

After years of offering best-in-class on-premises Real-Time transcription, we’re exceptionally proud to launch our Real-Time SaaS offering. This low-risk, high-reward approach to speech-to-text offers a perfect balance of fast results and highly accurate output, deployed within a secure public cloud environment.

This release will open up our technology to every size of business, giving more users access to the most accurate and fast speech-to-text currently available. You can see for yourself how easy it is to use and integrate by accessing our portal.

And Finally

Last but not least, we’ve sped up our Batch SaaS, reducing turnaround time by up to 75%. Typically transcribing 60 minutes of audio now takes under 5 minutes for Standard and 10 minutes for Enhanced.

For more details on all of our updates, you can find release notes here. If you need any additional support on these or any of the above, please contact our Support team.

Paul Gordon, Product Marketing Manager, Speechmatics