A robot playing piano
A robot playing piano. Image via Unsplash.

Limitations of Text-to-Speech

Written by Kirey Ismaya on February 21, 2021

TTS allows your device to read Aloud to you from digital text. TTS is the ideal tool to assist individuals with reading difficulties or to label specific items on the screen. TTS is used for people who are just learning to read or students who need extra support. Although TTS can read text clearly, it is clearly a synthetic voice or conversational AI without emotion or intonation; this is not a narrative or a show. Humans have excellent speech awareness and most of us can choose synthetic sounds when used. Intelligence can convey facts, information, and reminders, but it is not very adept at depicting subtext or emotion

A number of studies have highlighted various problems related to the text to speech system. One of the fundamental limitations of text to speech is producing emphasis and correct pronunciation from text input. The written text is emotionless and often can't pronounce the "name" correctly. In producing female or young children's voice characters, text to speech often has problems because women's voices have a tone that is almost twice as high as men's voices, and in the case of children it is even up to three times higher. The localization estimation of formant frequencies is more difficult with higher fundamental frequencies. There are also a number of problems associated with preprocessing text containing numbers, abbreviations, and acronyms.

1. Emotion

One of the limitations of text to speech is the inability to translate emotions. This is one of the potential limitations for the use of text to speech.

In the use of text to speech for entertainment purposes, the majority of the sounds produced by text to speech will sound much flatter when compared to the voice characters produced by a voice actor.

Adding emotion to the text to speech system requires certain speech segments to sound more pleasant, gentle or impolite, according to the communication situation. To achieve this goal, two problems must be solved: emotions must be identified on the basis of the input text, and a corresponding signal change must occur during the production of synthetic speech.

2. Prosody

The resulting prosody is one of the main problems faced by the text to speech system. The problem in question can come from the prosodic basis (speech with a little emotion) of expression and nuances in the delivery.

When humans read out a text, listeners gather contextual information. The prosody of a sentence is usually determined by the information presented in the previous few sentences. However, the current TTS system cannot make good use of this information. Therefore, text to speech seems to have a rhythm and dynamics that are not good when compared to the human voice.

3. Naturalness

The sound quality that seems natural is one of the determinants of the quality of a text to speech system. Even though in more modern times like now, the sound produced by text to speech does not completely remind us of robotic voices and is heard more by authentic human voices, there are still some degradations that reduce the impression of the quality of a text to speech system.

4. Ambiguity

Ambiguity is often a problem in the text to speech systems. One of the most common problems is ambiguity in a homograph or two words that have different meanings but have the same written form. Ambiguity in text to speech can affect the quality of the voice produced from the converted text. This problem is also often found when the text to be converted using text to speech contains a "name".

The flat voice and the limitation to convey emotion is the main reason why using text to speech for video applications, video game dialogues, or audiobooks is not the best choice. When performance is needed or emotions need to be conveyed, the human narrator is an irreplaceable choice.

For defining words, creating initial videos during production, and providing service or support, TTS is a perfect fit. In education, short segments of an artificial narrative can be very self- explanatory.

Conclusions

In using the TTS system, everyone naturally has their own tastes starting from the character of the voice or how natural the sound produced by the text to speech device is. Whether for learning or entertainment, you can try the text to speech feature from Aloud which provides a more natural voice and doesn't sound like a robot and provides a variety of voice characters.