With more people reading captions rather than listening to content, subtitles are big business. Now, AI looms over this job
This reflects in the global captioning and subtitles market. In 2021, it was just under $300 million, according to the data tracker Valuates Reports, and is expected to touch nearly $500 million by 2028. With so much demand, not only is there a surge in the number of subtitle writers and agencies supplying these services in the country, but also a new factor in the mix: AI.
Dubverse.ai is a Gurugram-based startup which uses generative AI to help users dub and subtitle their videos into over 90 languages, Indian and otherwise. While the startup was born to cater to edtech clients during the pandemic, today its clients include individual content creators, media houses, and streaming platforms.
Co-founder Anuja Dhawan says the main advantage is cost. “While dubbing is an old industry, it was very restricted to high-value clients such as filmmakers. The costs for dubbing are as high as Rs 2,000 per minute. However, with AI, we’ve been able to bring down the cost massively to as low as Rs 50 per minute,” says Dhawan. Since its launch, 5,00,000 users have signed up on the platform, Dhawan claims, adding that nearly 30% of them are from outside India. It is also in the process of demoing its products to production houses such as Sony.
In order to cut costs, production houses and streaming platforms are also turning to the use of artificial intelligence. Eros Now, for instance, tied up with Google Cloud to use AI technology for streaming its entire range of movies and originals with automated subtitles.
Rohit Shivdasani, co-founder of post-production services provider Atom Ant Studios which pivoted to subtitling in the pandemic, says the sheer volume of demand for subtitles makes the use of AI inevitable. “The demand is coming not only from new content, but the hundreds of shows and films that are being acquired by streaming platforms and added to their library,” says Shivdasani. “For live broadcasting, the subtitles need to be churned out in multiple languages on the same day. It’s just not possible for humans to do it.”
Another player is Mumbai-based startup Banva — named after the Marathi word for creating. It not only provides AI-based subtitle generation services, but also a tool that allows users to edit the generated subtitles quickly. About 2,000 people have already signed up, its founder Rahul Pawar says. “While writing AI-generated subtitles for the field of entertainment is tricky due to local nuances, sarcasm and idioms, clients are realising that a software like this can lead to much faster turnaround times and reduce the manual workload by almost 80%,” he says.
Banva’s pricing today costs Rs 500 for generating subtitles for a one-hour long video, and a credit-based system for longer duration needs. Dubverse, on the other hand, has monthly subscriptions starting from Rs 3,000. In contrast, the average rates charged by a manual subtitle writer range from Rs 80-100 per minute.
Despite the promise that the world of subtitling holds for AI-based firms, there are considerable challenges in understanding, translating, and subtitling the vast pool of Indian languages. Obstacles include the gendered nature of objects, which is absent in English, as well as accents and cultural contexts. Pawar says the platform is trying to negotiate these hurdles. “A lot of content in India is multilingual. Different characters are speaking in Hindi, English, Marathi, and so on. To handle that, we are training our models to better handle multilingual speech.” Yet, the accuracy levels differ for different language pairings, he admits. For instance, while English subtitles generated for Hindi content are nearly 90% accurate, the accuracy is lower for other languages. For subtitles from one regional language to another, the road to success is longer.
The question plaguing the subtitling and dubbing industry is the same one that is at the centre of the writers’ strike in Hollywood: Will AI kill human workers in the industry? Veteran subtitle writers like Vivek Ranjit are confident that writing good subtitles is an art and cannot be replaced by technology. “We can’t predict how far AI tech will progress, but I believe that we will always need a human touch, especially in a country like India where we have so many different cultures. Things keep changing from state to state and we have songs with expressions that cannot be literally translated,” says Ranjit, who has written subtitles for over 220 films. AI players, on the other hand, say the tech was neither meant to open the doors to a battle between writers and computers nor was it meant to have 100% accuracy. Banva’s Pawar says, “We are not trying to take away jobs from subtitlers. We are just trying to make their jobs easier.”
This reflects in the global captioning and subtitles market. In 2021, it was just under $300 million, according to the data tracker Valuates Reports, and is expected to touch nearly $500 million by 2028. With so much demand, not only is there a surge in the number of subtitle writers and agencies supplying these services in the country, but also a new factor in the mix: AI.
Dubverse.ai is a Gurugram-based startup which uses generative AI to help users dub and subtitle their videos into over 90 languages, Indian and otherwise. While the startup was born to cater to edtech clients during the pandemic, today its clients include individual content creators, media houses, and streaming platforms.
Co-founder Anuja Dhawan says the main advantage is cost. “While dubbing is an old industry, it was very restricted to high-value clients such as filmmakers. The costs for dubbing are as high as Rs 2,000 per minute. However, with AI, we’ve been able to bring down the cost massively to as low as Rs 50 per minute,” says Dhawan. Since its launch, 5,00,000 users have signed up on the platform, Dhawan claims, adding that nearly 30% of them are from outside India. It is also in the process of demoing its products to production houses such as Sony.
In order to cut costs, production houses and streaming platforms are also turning to the use of artificial intelligence. Eros Now, for instance, tied up with Google Cloud to use AI technology for streaming its entire range of movies and originals with automated subtitles.
Rohit Shivdasani, co-founder of post-production services provider Atom Ant Studios which pivoted to subtitling in the pandemic, says the sheer volume of demand for subtitles makes the use of AI inevitable. “The demand is coming not only from new content, but the hundreds of shows and films that are being acquired by streaming platforms and added to their library,” says Shivdasani. “For live broadcasting, the subtitles need to be churned out in multiple languages on the same day. It’s just not possible for humans to do it.”
Another player is Mumbai-based startup Banva — named after the Marathi word for creating. It not only provides AI-based subtitle generation services, but also a tool that allows users to edit the generated subtitles quickly. About 2,000 people have already signed up, its founder Rahul Pawar says. “While writing AI-generated subtitles for the field of entertainment is tricky due to local nuances, sarcasm and idioms, clients are realising that a software like this can lead to much faster turnaround times and reduce the manual workload by almost 80%,” he says.
Banva’s pricing today costs Rs 500 for generating subtitles for a one-hour long video, and a credit-based system for longer duration needs. Dubverse, on the other hand, has monthly subscriptions starting from Rs 3,000. In contrast, the average rates charged by a manual subtitle writer range from Rs 80-100 per minute.
Despite the promise that the world of subtitling holds for AI-based firms, there are considerable challenges in understanding, translating, and subtitling the vast pool of Indian languages. Obstacles include the gendered nature of objects, which is absent in English, as well as accents and cultural contexts. Pawar says the platform is trying to negotiate these hurdles. “A lot of content in India is multilingual. Different characters are speaking in Hindi, English, Marathi, and so on. To handle that, we are training our models to better handle multilingual speech.” Yet, the accuracy levels differ for different language pairings, he admits. For instance, while English subtitles generated for Hindi content are nearly 90% accurate, the accuracy is lower for other languages. For subtitles from one regional language to another, the road to success is longer.
The question plaguing the subtitling and dubbing industry is the same one that is at the centre of the writers’ strike in Hollywood: Will AI kill human workers in the industry? Veteran subtitle writers like Vivek Ranjit are confident that writing good subtitles is an art and cannot be replaced by technology. “We can’t predict how far AI tech will progress, but I believe that we will always need a human touch, especially in a country like India where we have so many different cultures. Things keep changing from state to state and we have songs with expressions that cannot be literally translated,” says Ranjit, who has written subtitles for over 220 films. AI players, on the other hand, say the tech was neither meant to open the doors to a battle between writers and computers nor was it meant to have 100% accuracy. Banva’s Pawar says, “We are not trying to take away jobs from subtitlers. We are just trying to make their jobs easier.”