Meta, the company formerly known as Facebook, has made a significant leap in the field of artificial intelligence (AI). The company’s researchers have developed new AI models that can recognize and produce speech in more than 1,000 languages, a significant increase from the approximately 100 languages covered by existing speech recognition models.
Overcoming the Language Barrier
There are around 7,000 languages spoken worldwide, but the majority of them lack comprehensive coverage in existing speech recognition models. This is primarily due to the requirement of large amounts of labeled training data, which is only available for a handful of languages like English, Spanish, and Chinese. To overcome this challenge, Meta researchers retrained an existing AI model developed by the company in 2020. This model can learn speech patterns from audio without needing large amounts of labeled data.
Training the AI Models
The researchers trained the AI model on two new datasets. The first dataset contained audio recordings of the New Testament Bible and its corresponding text in 1,107 languages. The second dataset consisted of unlabeled New Testament audio recordings in 3,809 languages. The team processed the speech audio and text data to improve its quality before running an algorithm designed to align audio recordings with accompanying text. This process was repeated with a second algorithm trained on the newly aligned data. This method enabled the researchers to teach the algorithm to learn a new language more easily, even without the accompanying text.
Performance and Limitations
Meta’s researchers claim that their models can converse in over 1,000 languages and recognize more than 4,000. They compared their models with those from rival companies, including OpenAI Whisper, and claim theirs had half the error rate, despite covering 11 times more languages. However, the team also warns that the model is still at risk of mistranscribing certain words or phrases, which could result in inaccurate or potentially offensive labels. They also found that their speech recognition models yielded more biased words than other models, albeit only 0.7% more.
The Controversy of Using Religious Texts
While the scope of the research is impressive, the use of religious texts to train AI models can be controversial. Chris Emezue, a researcher at Masakhane, an organization working on natural-language processing for African languages, who was not involved in the project, points out that “The Bible has a lot of bias and misrepresentations.”
Conclusion
Meta’s breakthrough in AI speech recognition and production for over 1,000 languages is a significant step forward in making AI more accessible and useful to people around the world. However, the challenges and controversies surrounding the training data and potential biases in the models highlight the complexities and ethical considerations in the development of AI technologies