Visual feedback was found to be effective in improving production of non-native vowels (Kartushina et al., 2015). However, subjects had problems producing certain target vowels that were perceived to be in the same category in their native language. There was also an overall improvement in non-native vowel perception due to the production training, which heightened awareness on how these vowels are produced. This supports the Motor Theory of speech perception, which argues that listeners use their implicit knowledge of how a sound is produced to perceive that sound (Liberman & Mattingly, 1985).

While Automatic Speech Recognition can help learners to practice speech production on their own, this method was not as effective as expected. This is possibly because artificial intelligence is not sophisticated enough to capture the full acceptable range of variable sounds and often fails to recognize words in the second language that are actually correct (van Doremalen et al., 2016). Therefore, this training method is promising, but the software will have to be refined. Furthermore, it only provides feedback on whether the pronunciation is correct or incorrect, and is not as helpful as visual feedback training, as learners will not know how to self-correct their speech production and will be required to guess until the software approves their pronunciation.

For perceptual training, the experimental group was successful in seeing positive outcomes for the word identification task, and subjects were able to assimilate what they had learned from the training and apply the distinctions to new words as well. Improved perception of the target sounds also led to enhanced production of the same sounds, suggesting that speech perception and production are related, although they do not have a one-to-one relationship.

On the other hand, HVPT was found to significantly improve native Mandarin speakers’ capacity to identify L2 English vowels, as the speed of identification was increased following the training, compared to learning contexts where learners are not as conscious of highly variable input (Thomson, 2012). Furthermore, HVPT conclusively provides more accurate feedback than ASR, and has been shown to boost both L2 speech production and perception, even though it only trains speech perception.

For L2 speech production, visual feedback has been found to be more effective than automatic speech recognition in aiding second language learners. The visual feedback method also slightly improved the ability of learners to distinguish the L2 sounds that they could better produce. On the other hand, the perceptual training method that was most successful, proved to be HVPT.