The Cloud Text-to-Speech service, which uses neural networks to convert text to speech, can be used by developers in their apps. Man...
The Cloud Text-to-Speech service, which uses neural networks to convert text to speech, can be used by developers in their apps.
Many Google products, including the personal assistant available on smartphones, use voice synthesis to generate a natural voice. As developers have asked to integrate the technology into their applications, the Mountain View company has decided to release the Cloud Text-to-Speech service through the Google Cloud Platform.
Cloud Text-to-Speech can be used in a variety of ways, for example, for automated call center answering systems (IVR), so as to offer a conversation in natural language and in real time. The service can also be used to give voice to IoT devices (TV, cars or robots) and to convert articles or books into podcasts or audiobooks. Users can choose from 32 entries in 12 different languages (others will be added later). Cloud Text-to-Speech correctly pronounces complex text, such as names, dates, times and addresses.
You can change the pitch, speed and volume of the voice, and generate the output in various audio formats, including MP3, WAV and Opus. The cost of standard items (after the first 4 million characters) is 4 dollars for every million characters. Those looking for higher quality can choose WaveNet voices generated by the artificial intelligence developed by DeepMind, a division of Alphabet. In this case the price rises to 16 dollars per million characters, after the first million free.
WaveNet is the name of the neural network trained with a huge number of samples in order to obtain vocal waveforms practically indistinguishable from human ones. The new WaveNet version, based on the Cloud TPU infrastructure, offers a 1,000 times higher speed than the original model and generates waveforms with 24,000 samples per second. The audio quality has also been increased bringing the resolution of each 16-bit sample.
The voices generated by artificial intelligence are very similar to human ones and 20% better than those obtained with other systems. The quality will be further increased over the next few months. Among the first companies that use Cloud Text-to-Speech are Cisco and Dolphin ONE.
