Bengaluru: Bringing Amitabh Bachchan’s voice to Alexa involves two major technology challenges for Amazon.
The voice must sound exactly like Bachchan, because it is a voice that only recognizes Indians well.
As Manoj Sindhwani, Vice President of Alexa’s speech on Amazon, said, “My mother fans of Bachchan.
I am worried if there is one weakness, I will not hear it.” It’s more complicated, said Sindhwani, by the way where Bachchan spoke – it was very rich, and it was very rich, and He spoke with many emotions, intonation.
It’s difficult for the text-to-greeting sound system to perfect.
The second big challenge is to use ‘Amit Ji’ as Wake said.
Wake Word is the word you use to activate Alexa – which has so far been ‘Alexa’.
The Amazon team considers other wake-up words like ‘Mr.
Bachchan’, ‘Bachchan Ji’, ‘Amitabh Bachchan Ji’, ‘Amitabh Ji’.
But nothing sounds like pulling it with ‘Amit Ji’.
But it’s very short, practically one syllable, that many other words we use in the sound of everyday greetings are similar to it.
You can even have older people in your home which are amit or ajit.
Will be annoying to have Alexa often wake up with things that shouldn’t.
We will know how well Amazon solves this problem only when people start using ‘Amit Ji’ on a large scale.
In Amazon, they are happy with what they have achieved.
Bachchan is just the fourth celebrity, and the first is outside the US, to be part of Alexa’s voice feature.
The first celebrity voice used was American actor Samuel LL Jackson, which was launched in December 2019.
The work with Bachchan involved a technology team in India, Poland, England and the US, and the actor recorded his voice in many sessions, so that the artificial intelligence system (AI) can then Do it.
Puneesh Kumar, country leader for Alexa on Amazon India, said – for all entertainment – that a voice engineer in Poland was on the basis of the first name with Bachchan, given all the interactions they had.
Bachchan, he said, is Sticker for the standard.
“There are so many opportunities where we feel, oh, this sounds pretty good, close enough to your voice.
And he’s like, no, let’s try again, I want to get it perfect.” The dominant technology used to enhance Bachchan’s speech is called a system.
Text-to-speech nerve.
When you ask questions, the system first turns it into a text, look for the answer, and then change the answer from the text into the bachchan sound.
“There are several ways to do text-to-speech, but the latest and largest are based on deep neural networks or deep learning,” said Sindhwani.
This is one of the most sophisticated forms of learning or machine.
“This training method is able to produce a model that not only reproduces Mr.
Bachchan’s voice, but also the style of speaking – the way he emphasizes certain words, quickly on certain occasions, slowing down some.
Many innovations and thinking goes into this,” he said.
Another complication is to ensure Alexa will recognize ‘Alexa’ and `Amit Ji ‘as the time.
It will usually take too much memory and calculation.
“So, we use what we call multi-target learning, where you have one input, and you try to predict some output.
It’s very complex, requires a lot of thought about how we build a model.
And on that, it’s covid, I don’t Can collect a lot of data, but must work for a unique environment in India, with all the noise that usually exists, “said Sindhwani.
To overcome the lack of data, Amazon uses what is called Transfer Learning, where you take the skills learned from one domain and transferring different learning.
“We transfer learning from something that works for the recognition of moderate vocabulary, for very specific recognition, which is extraordinary,” said Sindhwani.