Diffio web app restores speech in noisy interview recordings

Diffio is a web application that restores speech in uploaded audio and video recordings. It was created by Nathan Harmon and is offered by Ozni Apps LLC. The target material is an interview recorded somewhere noisy, a distorted phone call or archival audio that needs a boost.

The service works on a recording after it has been captured. Diffio uses diffusion neural networks to pull speech away from background sound. A user uploads a file in the browser, and processing runs in the cloud.

For radio and television producers, that covers material recorded outside their own microphone setup, such as a viewer-submitted video or a remote interview. Diffio complements microphone placement and recording practice and does not replace them.

What the player offers

An in-browser player lets an editor compare the original with the restored speech and adjust how much background sound stays in the mix. Restored speech, background sound and a finished mix can be downloaded separately. A video file keeps its picture, and only the speech soundtrack is restored.

Two model options are available, Diffio 4.5 Flash and Pro. The company's product page says to start with Flash for a decent room with a little hum, and to use Pro for a guest recorded on a laptop in a loud space. A developer API lets other workflows call the speech restoration.

Harmon says the tool is built for speech rather than music. "Crowd voices and other speech-like sounds can sometimes be assigned inconsistently between the speech and background outputs," he says. Results depend on the source recording, so editors should compare the output with the original before broadcast or publication. "A cleaner-sounding result does not by itself establish that words have been preserved accurately," he notes.

Diffio is independently funded. It grew out of work on restoring older sermon recordings and now covers spoken-word audio and video.

Source: Diffio