When Alexa talks, it’s using text-to-speech. Niclas Bergström join us to lift the curtain and tell us what’s actually happening under the hood when voice assistants talk.
Every voice assistant needs three core components: Automatic Speech Recognition (ASR), Natural Language Understanding (NLU) and Text-to-Speech (TTS). We’ve already covered what Automatic Speech Recognition is and how it works with Catherine Breslin and in this episode, we’re covering the latter, text-to-speech.
Text-to-speech is the technology that gives voice assistants a voice. It’s the thing that produces the synthetic vocal sound that’s played from your smart speaker or phone whenever Alexa or Siri speaks. It’s the only part of a voice assistant that you’d recognise. The other core components, ASR and NLU, are silent.
And, given how we’re hard wired for speech – a baby can recognise its mother’s voice from the womb – how your voice assistant or voice user interface (VUI) sounds is one of the most important parts of it.
A voice communicates so much information without us necessarily being aware. Just from the sound of someone’s voice, you can infer gender, age, mood, education, place of birth and social status. From the sound of someone’s voice, you can decide whether you trust them.
With voice assistants, voice user interfaces, or any hardware or software that speaks, choosing the right voice is imperative.
Some companies decide on a stock voice. One of Readspeaker’s 90 voices or perhaps Amazon Polly. Others create their own bespoke voice that’s fit for their brand.
We see examples of Lyrbird’s voice cloning and we hear Alexa speak every day, so it’s easy to take talking computers for granted. Because speaking is natural and easy for us, we assume that it’s natural and easy for machines to talk. But it isn’t.
So in this episode, we’re going to lift the curtain on text-to-speech and find out just exactly how it works. We’ll look at what’s happening under the hood when voice assistants talk and see what goes into creating a TTS system.
To guide us through the ins and outs of TTS, we’re joined by Niclas Bergström, a TTS veteran and co-founder of one of the largest TTS companies on the planet, Readspeaker.
Readspeaker is a pioneering voice technology company that provides lifelike Text to Speech (TTS) services for IVR systems, voice applications, automobiles, robots, public service announcement systems, websites or anywhere else. It’s been in the TTS game for over 20 years and has in-depth knowledge and experience in AI and Deep Neural Networks, which they put to work in creating custom TTS voices for the world’s biggest brands.
Visit Readspeaker.com to find out more about TTS services
And Readspeaker.ai for more information on TTS research and samples