CAMB.AI Creates Breakthrough Voice AI Using High-Performance GPUs from AWS

Founded in 2023, CAMB.AI is an advanced AI speech synthesis and translation company. Its proprietary speech-to-speech translation models are used by hundreds of enterprises for multilingual content creation. CAMB.AI supports a wide range of applications, including movie production, live events, and multinational news broadcasts. “Our technology transforms audio/video content, whether live or prerecorded, from its source language into over 140 languages and dialects,” says Akshat Prakash, chief technology officer at CAMB.AI. “We do all of this while maintaining the tonality, voice, and performance of the original speakers or actors.”

CAMB.AI needed to develop and train sophisticated AI models for speech synthesis and translation across a vast number of languages, which presented major computational challenges. These models require extensive processing power and memory, which, in turn, necessitates access to a large number of high-performance GPUs. As a startup, CAMB.AI initially struggled to find a cloud provider willing to support its high-resource needs and help bring this vision to life.

After consulting with different providers, CAMB.AI found one that was ready to step up to the challenge: AWS. “We knocked on the doors of many cloud providers in the past who were unable to support our ambition and needs,” says Prakash. “Perhaps other cloud providers did not believe that we could create a breakthrough speech-to-speech translation model. AWS did, backed us all the way, and they provided us with more support than anyone else out there.”

CAMB.AI adopted Amazon Elastic Compute Cloud (Amazon EC2) P5 Instances, which are high-performance GPU-based instances for deep learning and high-performance computing applications. These GPUs provided the processing power and flexibility for training the startup’s sophisticated speech and translation models. Using AWS infrastructure, CAMB.AI efficiently trained and deployed a new version of its speech model, MARS5, which received over 2,000 stars on GitHub and continues to be a landmark model in the open-source world of voice AI.

“We made an open-source version of MARS5, the fifth iteration of MARS, available in English to the public,” says Prakash. “It trended at number five on the Hugging Face leaderboard and got ratification from Mozilla AI, IBM, Carnegie Mellon, the Massachusetts Institute of Technology, and more.”

Using AWS, CAMB.AI can develop models more rapidly than ever before. In fact, the startup trained MARS5 using Amazon EC2 P5 Instances over the course of 35 days before the complex model was ready for deployment—a process that would have taken four times as long without AWS services. With these training capabilities, CAMB.AI can offer live, accurate AI dubbing in over 150 languages to its customers quickly and at scale, a feat that other cloud providers had considered unattainable.

CAMB.AI also adopted services like Amazon Simple Storage Service (Amazon S3)—which is built to retrieve any amount of data from anywhere—to manage and process the vast amounts of audio data required for training its models. Additionally, completing all processes, from data storage to pipelining and training, on AWS as opposed to working with multiple providers provided unparalleled cost savings up to ten times less what CAMB.AI remotely received as offers from even tier-two providers.

Using AWS, CAMB.AI has the confidence to deploy its solutions worldwide, knowing it can scale effortlessly to meet growing demand across different regions. The startup can handle complex, large-scale speech synthesis and translation tasks, including near real-time dubbing for live events, processing massive volumes of content for global media companies, and providing instant translation services for multinational corporations—all while maintaining the nuanced qualities of the original audio.

Looking to the future, CAMB.AI is participating in the AWS Generative AI Accelerator 2024, a global 10-week hybrid program supporting generative AI startups to prove what’s possible with an endowment of $1,000,000 or more credits. This program will empower CAMB.AI to further push the boundaries of AI-driven language technology.

Next, the startup is looking to take advantage of global AWS infrastructure to provide high availability and low latency to its clients. By achieving high availability on AWS, the company can better serve global traffic and support live streamed events that have stringent service-level agreements.

“The support, understanding, and availability that AWS offered CAMB.AI was unparalleled,” says Prakash. “The AWS Generative AI Accelerator will significantly help us in our journey toward becoming the first true, homegrown AI unicorn startup out of Dubai.”