Voice Generating Startup Resemble AI Promises to Restore Old Audio
While many AI companies race to find ways to use the technology to enhance or even create video, Resemble AI is focused on audio fidelity. The startup—which also offers an AI voice generator for businesses to use to create realistic human–like voiceovers—has launched ‘Resemble Enhance,’ an open-source tool designed to significantly upgrade the quality of historical audio.
The new service can take a distorted, fuzzy recording of a long-lost historical speech and then apply AI to make it sound like it was recorded or broadcast yesterday.
The Canadian company says Resemble Enhance is distinguished by its dual-module approach, combining a sophisticated denoiser—which removes static background hums and hisses—and an AI-powered speech enhancer. This combination not only removes unwanted noise but also enriches the overall quality of the audio.
🎙️Today, we introduce Resemble Enhance🚀 — our latest AI-powered model! Enhance is an open-source speech enhancement model that transforms noisy audio into noteworthy speech!
Read the technical blog: https://t.co/RCp4xi4GqR👩💻#AIVOICE #podcasting #audioengineer pic.twitter.com/kfdGmKlSgB
— Resemble AI (@resembleai) December 14, 2023
Even though there are other audio restoration products on the market, Resemble’s combination of techniques could be a meaningful differentiator.
How does it work?
The Resemble AI denoiser uses UNet, an AI model that helps to separate the different types of sounds that appear on a recording. It excels at filtering out unwanted noise from audio tracks, leaving just the speech as the focus.
Once UNet does its job, the enhancer module kicks in, extending audio bandwidth and correcting distortions. This dual functionality, the company says, ensures that the final output is not just noise-free but also possesses the richness of contemporary recordings.
As an open-source tool, Resemble Enhance is accessible at no cost, a compelling option in the traditionally expensive market of media restoration services. The primary beneficiaries of Resemble Enhance are industries reliant on clear audio quality, such as podcasting, entertainment, and education. Additionally, this tool offers a new lease on life to historical recordings, potentially providing clearer insights into the past.
The tool’s release comes when the demand for high-quality digital content is at an all-time high. Meanwhile, the open-source nature of Resemble Enhance positions it as a potentially disruptive force in a market currently dominated by high-cost proprietary solutions.
The convergence of AI in audio and video enhancement will likely pave the way for more comprehensive media restoration solutions.
By combining this tool with other video enhancers that use generative AI or other models to upscale and enhance images and faces—like GPEN or the well-known GFPGan—users can now achieve professional results with their own computers for a minimum investment.
To experiment with Resemble Enhance, users can visit the official Resemble AI website or download their models from the project’s official Github page.
Edited by Ryan Ozawa.