Google Cloud has announced that it is updating its text-to-speech products with more votes and more languages. It also improves the quality of its transcription tools and makes some of these functions widely available.
For Text-to-Speech, Google has almost doubled the number of votes available since the update in August. Support has also been added for seven new languages or variants, including Danish, Portuguese, Russian, Polish, Slovak, Ukrainian and Norwegian BokmÃ¥l. All these languages are still in a beta version. A total of 21 languages are now supported.
For those new languages, Google has added 31 new WaveNet voices and 24 new standard voices. In its own words, 106 votes are now supported. WaveNet is a deep neural network for generating raw audio, creating voices that sound more natural than standard text-to-speech voices. The technology was developed by DeepMind, the AI company that acquired Google in 2014.
“Thanks to the unique access to WaveNet technology driven by Google Cloud TPUs, we can now build new voices and languages faster and easier than is normal in the industry,” said product manager Dan Aharon. The main competition for text-to-speech services is AWS’s Polly. Polly now supports 58 voices.
Google has also now made the Device Profiles function generally available for Speech-To-Text. This allows users to optimize audio playback on various types of hardware. Think of headphones, where the audio is optimized for applications such as podcasts.
Premium models for video and enhanced phone are also made widely available. The video model is based on technology similar to what YouTube uses for automatic subtitling. According to Google, the technology now has 64 percent fewer transcription errors. The enhanced phone model has 62 percent less.
The models could be improved by requiring customers using premium services to share user data via data logging. From now on, users can use the enhanced phone-model without sharing data, although they have to pay more.This news article was automatically translated from Dutch to give Techzine.eu a head start. All news articles after September 1, 2019 are written in native English and NOT translated. All our background stories are written in native English as well. For more information read our launch article.