In 2021, voice artist Chetan Shashital was asked to voice Bal Thackeray for the Marathi film Balkadu, which is based on the Shiv Sena founder’s ideology. When Shashital send the producers a sample based on a speech that had been shared with him, pat came the reply: you have sent us back Thackeray’s speech by mistake.
Balkadu’s makers could not distinguish between Thackeray’s original voice and Shashital’s pitch-perfect imitation. Though most people don’t realise it, the talents of dubbing and voiceover artists like Shashital are part of our daily lives. We hear them not just in films and web shows dubbed into other languages, animated productions and corporate videos, but also on automated phone messages when we’re told to hold or press a button, in public service announcements and in car navigation systems.
But now, this thriving industry faces an unforeseen challenge: artificial intelligence.
Just like robots have supplemented workers in factories, AI software that can replicate human voices is edging artists out of jobs. In India, AI-aided voicing is already being used in corporate films, audiobooks and commercials, industry insiders say.
The situation is alarming enough to merit a pushback from the Mumbai-based Association of Voice Artists, whose 1,000 or so members work in advertising, film and television. “If AI takes over, we are finished,” said Amarinder Singh Sodhi, the association’s general secretary. “As voice artists, we need to get some regulation in place so that our livelihood is protected.”
‘Non-human competition’
When the Association of Voice Artists was established in 1999 to push for fair wages, settle payment disputes and arrange medical benefits, its members never anticipated that they would face a challenge from non-human competition.
Ganesh Divekar, the association’s president, explained how jobs for voice-over artists could evaporate with AI. A film that has 10 major characters and several minor ones would usually need 30 or 40 artists when it is being dubbed into another language, he said. But with AI, it could be done with two male and two female artists.
“AI can take the texture of the actors and apply it to the performances, a process that is called voice cloning,” he told Scroll.
The process of harvesting voices to be fed into AI programmes has been underway for at least a decade, the association’s office-bearers say. “Voice artists were giving voice samples to technology companies without realising that these samples would be used to generate new content that could be exploited till eternity,” Sodhi said. “They were paid what they thought was a handsome sum, but they ended up signing over their voices for life.”
In one instance, a voice artist told the association about being hired by an automotive company for what she was thought was a vehicular navigation system. “Without any contract in place, she started recording her voice,” said Akshay Shetty, the association’s lawyer. “But the script that was given to her had nothing to do with navigation. She had to read random words and excerpts from a novel. What she didn’t realise than was that her voice was being used to train the AI.”
While AI sites provide tools to generate voices, the software has been constructed using human inputs, Shetty noted. “Somebody has fed a particular voice into the algorithm,” he said. “So should that particular voice artist not be paid for the new content that is being generated?”
Vanishing work
The voice artists realised the extent of the challenge they faced when assignments started disappearing, Shetty added. “The guys who were giving voices for advertisements and corporate audio-visuals found that the work was drying up,” he said. “Why would that be the case when these sectors were actually growing?”
The union has issued guidelines to help its members navigate this unfamiliar terrain. “We can’t stop the march of technology,” Shetty said. “What we are saying is, know what you are getting into. The best we can do is create awareness. If artists do decide to turn over their voice rights to AI companies, they should charge them well since this it’s for perpetuity.”
Sodhi added: “We’re telling our members to be clear about what exactly will the voice be used for, for how long, for how many mediums.” He said that this is “a survival of a fittest scenario, so we have to negotiate options for everybody concerned”. He added that if even one of the association’s members benefits from these guidelines, it would be a victory for all.
India’s secret weapon
The AI problem does not hurt only voiceover and dubbing artists but also the dubbing producers who hire them for assignments. AI is attractive to producers looking to cut costs, even if the AI-generated voice sounds mechanical. “If a streaming platform tells a producer that it needs just a couple of voices and will use AI to deal with the rest, the dubbing producer’s revenue is immediately halved,” Sodhi pointed out.
Two factors have kept AI away from Indian films and shows so far. One is that AI-generated voices sound tinny and machine-like. “AI hasn’t impacted voicing in films because it doesn’t give the desired results,” said dubbing producer Ashim Samanta.
He recalls being sent samples by an Israeli company that had dubbed a series with AI software. “It was so bad that I couldn’t stand it beyond two minutes,” Samanta said. “Dubbing requires drama, emotions and expression. Whatever we have seen so far is pathetic.”
But he added, “ Of course, we don’t know what will happen a few years from now.”
Ganesh Divekar used a food analogy to compare the benefits of humans over machines: “You can make chutney in a mixer or with the traditional method. The taste is completely different. That is the difference. AI can’t pull it off in terms of performance just yet. AI needs a command. I do not need a command. What I do is with my experience and expertise.”
The Indian accent
The bigger hurdle for AI is already evident in navigational tools and virtual assistants: the Indian accent is impossible to crack because there isn’t a standard Indian accent.
India’s staggering linguistic diversity, range of accents and colloquialisms have created a high wall between the voice industry and AI that will be very hard to scale, at least for some years. “When people ask me to send a voice sample, I ask, which one?” said Chetan Shashital, who has voiced numerous commercials and dubbed for actors across languages.
Voicing is a form of acting too, involving skilful mimicry as well as dramatic expression, Shashital pointed out. His first major film job was in 1989, when he dubbed for Pinchoo Kapoor after the actor died suddenly. Shashital was 18 years old; Kapoor 62. Shashital didn’t just have to sound like Kapoor but also emote in the actor’s style.
“We have the intelligence, whereas AI is artificial – how will it generate the accents and emotions?” Shashital said. His contract threatens legal action against his voice being used by AI programmes.
Vijayan Menon, a voiceover artist, noted that while he could refuse to share his scratch voices or sample voices, struggling artists in the early stages of their careers might not be in a position to refuse the money.
“Some artists pick up a couple of thousand bucks by jumping from studio to studio,” Menon said. “They can easily be replaced by AI, which can dub incidental characters, such as the doctor who has a single line in a film that it’s too late and the patient can’t be saved.”
The human touch is missing – for now, said dubbing producer Eliza Lewis. “But the day AI cracks Indian emotions, it will get really difficult,” she added. “People adapt – 35mm was replaced by digital technology. The local grocer faced competition from the likes of DMart. The grocer is still around, but for how long?”