Hi! what is SANA you ask..? She is an automated text to speech AI that can read sentences or text with expression. SANA can understand the difference between "angry", "emotional" scenario in a given text which she will use to create that voice to read the text aloud.
Text to Speech generation nowadays are done via one singular tone. There are no expression and it sounds very robotic. SANA on the other hand is 100% robotic but the way she manipulates her voice to create "fake" emotion is what matters and makes it sound realistic.
Text to speech (TTS) is a natural language modeling process that requires changing units of text into units of speech for audio presentation. This is the opposite of speech to text, where a technology takes in spoken words and tries to accurately record them as text. Text to speech is now common in technologies that seek to render audio output from digital text to assist those who are unable to read, or for other kinds of uses.
Developing text-to-speech capability includes some unique challenges. Especially in the English language, where a great number of homonyms have varied pronunciations, computer programs rely on probability modeling to guess the desired pronunciation of a word in digital text. The program also has to convert units of text into phonemes, the smallest units of speech pronunciation. The result is that many text-to-speech technologies are less than infallible, although SANA will be expressible ;)
While text to speech has benefits for all users, some of the specific groups that see a better user experience are:
- People with learning disabilities
- People who have literacy difficulties
- People who speak the language but do not read it
- People who multitask
- People who access content on mobile devices
- People with different learning styles
- Also the fact that Lina puzzle hint 2: https://i.imgur.com/82w6TI7.png
We can evaluate the expression of SANA with MOS model. MOS gives a numerical indication of the quality of the synthesized speech. In this method, focus of evaluator should be on naturalness of synthetic speech. By Naturalness it is meant that the sound is indistinguishable from human speech and it is close to human voice.
S(i) Score of i'th evaluator N = # of Evaluators M = # of Sentences j = Sentence Index
Given the output of formula above where MOS > 0 and < 6, where 5 is Excellent expression and 1 being very robotic.