Digital Event Horizon
Mozilla's Common Voice project has made significant strides in creating a more inclusive dataset for voice-activated assistants. By collecting and contributing high-quality audio data in various languages, the project aims to democratize AI voices and promote linguistic diversity. However, concerns remain about the lack of control over how donated data might be used afterwards, highlighting the need for greater transparency and regulation in the tech sector.
The current lack of diversity in voice-activated assistant voices is a significant issue. The majority of default voices are white American or British, speaking English, due to limited data training. The Common Voice project aims to create a more inclusive and diverse dataset for voice-activated assistants. Volunteers have contributed over 31,000 hours of high-quality audio data across 180 languages. The project's permissive license allows companies like Mabel AI to use the dataset. Ensuring diversity, especially gender representation, is a key challenge in collecting voice data. The Common Voice project prioritizes socioeconomically diverse sets of speakers and aims to transmit cultural context.
The world of artificial intelligence (AI) is rapidly evolving, and one area where significant progress has been made is in the development of voice-activated assistants. These intelligent devices are becoming increasingly ubiquitous, with companies like Apple and OpenAI rolling out new generations of AI-powered assistants that promise to revolutionize the way we interact with technology. However, a glaring gap in this technological landscape remains: the lack of diversity in the voices used to power these assistants.
Currently, most default voices for voice-activated assistants are white American or British, speaking English. This limitation is not only a reflection of the dominance of the English language but also a result of the limited data that has gone into training AI models. The vast majority of this data comes from the English-language internet, which reflects Anglo-American culture.
To address this issue, Mozilla's Common Voice project was launched in 2017 with the aim of creating a more inclusive and diverse dataset for voice-activated assistants. This grassroots effort relies on volunteers to collect and contribute high-quality audio data in various languages, making it one of the most useful resources for building voice AI.
The project has seen significant growth over the past seven years, with a massive spike in downloads partly due to the current AI boom. The Common Voice dataset now boasts an impressive 5 million mark, up from 38,500 in 2020. Volunteers have contributed a total of 31,000 hours of voice data across around 180 languages, including Russian, Catalan, and Marathi.
The project's permissive license means that many companies rely on it, such as the Swedish startup Mabel AI, which built translation tools for healthcare providers using Ukrainian voices. The team has since expanded to other languages like Arabic and Russian.
One of the key challenges in collecting voice data is ensuring diversity, particularly when it comes to gender representation. Researchers have found that most voice data collected by volunteers is donated by younger men, which does not accurately reflect the demographics they are intended for. To address this issue, Common Voice asks participants to share their gender and accent details.
Rebecca Ryakitimbo, a Common Voice fellow who created the project's gender action plan, emphasizes the importance of collecting AI voice data from socioeconomically diverse sets of speakers. She has collected voice data in Kiswahili in Tanzania, Kenya, and the Democratic Republic of Congo, reaching out to women young and old living in rural areas who may not always be literate or have access to devices.
The Common Voice project is bolstered by the belief that languages form a crucial part of identity. EM Lewis-Jong, the product lead for Common Voice, notes that "it's not just about language, but about transmitting culture and heritage and treasuring people's particular cultural context." The team aims to create high-performance translation models that can convey information in a compelling way to people from different linguistic backgrounds.
Willie Agnew, a researcher at Carnegie Mellon University, praises Common Voice for its commitment to diversity. "I'm very impressed with how well they've done that and how well they've made this data set that is actually pretty diverse," he says. "It feels like they're way far ahead of almost all the other projects we looked at."
However, despite these efforts, concerns remain about the lack of control over how donated voice data might be used afterwards. The tech sector has a reputation for not always giving proper credit to volunteers, and the data is available for anyone's use.
In conclusion, Mozilla's Common Voice project represents a significant step towards democratizing AI voices and making them more inclusive. By relying on volunteers and leveraging grant funding, the project aims to create high-quality audio datasets that can help break down communication barriers and promote linguistic diversity. As the world of voice-activated assistants continues to evolve, it is essential that efforts like Common Voice are supported to ensure that these technologies serve diverse communities.
Related Information:
https://www.technologyreview.com/2024/11/15/1106935/how-this-grassroots-effort-could-make-ai-voices-more-diverse/
Published: Fri Nov 15 06:02:55 2024 by llama3.2 3B Q4_K_M
