Historical audio
How to Restore Old Recordings with AI
Ninety-year-old recordings don't have to sound like it. Hear what Amelia Earhart, a 1948 convention hall, and decades of magnetic tape sound like after Diffio.
Why Old Recordings Degrade
Every recording ever made was shaped by the technology available at the time, and that technology had strict limits.
Recordings from the 1930s through the 1950s were pressed onto lacquer transcription discs: 16-inch aluminum-core platters coated with nitrocellulose lacquer. As the Library of Congress has documented, these discs are inherently unstable. The lacquer coating chips and peels over time; surface noise accumulates with every playback; and the cutting and playback stylus system compressed dynamic range and cut off nearly all audio above 8–10 kHz. Of the 500,000 recorded radio broadcasts held by the Library of Congress, only about 50 come from the 1920s, and early 1930s materials face comparable rates of loss.
Magnetic tape, which became the dominant medium after World War II, brought its own vulnerabilities. Mold, brittle binders, shed oxide layers, and sticky shed syndrome (where tape absorbs moisture until playback physically destroys the original) threaten recordings from the 1950s through the 1980s. Consumer cassette tape, widely used for conference and lecture recordings in the 1970s through 1990s, pushed those limitations further: frequency response limited to roughly 50 Hz–12 kHz, prone to pitch instability, ambient hiss, and dropouts.
The result is that some of the most important voices in history (political leaders, scientists, aviators, labor organizers) are audible only through a wall of noise that modern listeners struggle to penetrate.
AI audio restoration changes that. Below, hear four of those voices restored. Each player compares an original capture with Diffio-enhanced audio using the same on-site demo pipeline as our homepage samples.
Amelia Earhart: 1930s Radio Recording
Before (Original) · After (Diffio Enhanced)
Amelia Earhart (1897–1937) was the first woman to fly solo across the Atlantic, a record-holder in speed and altitude, and one of the most widely recognized public figures of her era. She maintained an active radio presence throughout the 1930s, speaking on NBC and CBS about aviation, women's ambitions, and the future of flight, making her voice one of the most broadcast of any woman of that decade. She disappeared over the Pacific on July 2, 1937, three days before she was scheduled to appear on Lux Radio Theatre, leaving behind a handful of surviving recordings that are among the most historically significant audio documents in American history.
Her original recordings were captured on lacquer transcription discs under the standard conditions of 1930s radio: compressed dynamic range, persistent surface noise, and high-frequency content largely absent. Diffio removes the crackle and background hiss, pulls Earhart's voice out of the noise floor, and restores the clarity and presence of her speech without altering its character.
Alben Barkley: 1948 Democratic Convention Speech
Before (Original) · After (Diffio Enhanced)
Alben Barkley (1877–1956) served as the 35th Vice President of the United States under Harry S. Truman and spent four decades in the House and Senate, one of the longest continuous congressional careers in American history. He is remembered as one of the finest political orators of his era: warm, encyclopedic, and capable of moving an audience from laughter to outrage within a single speech. His keynote address to the 1948 Democratic National Convention, delivered at age 70 to a demoralized, fractured party in a cavernous Philadelphia convention hall, was so galvanizing that it spontaneously launched his vice-presidential candidacy. According to Smithsonian Magazine, delegates broke into an impromptu march singing My Old Kentucky Home and carrying Barkley for Vice-President banners; Truman called him the next morning to offer him the ticket.
The recording was made in a large, acoustically difficult convention hall, capturing a speech that ran hours behind schedule amid constant crowd noise. Diffio reduces the ambient crowd noise and disc surface artifacts, letting Barkley's voice, and the rhetorical force that changed an election, come through as intended.
Isaac Asimov: Science Lecture Recording
Before (Original) · After (Diffio Enhanced)
Isaac Asimov (1920–1992) published more than 500 books over his lifetime, spanning science fiction, chemistry, history, mathematics, and nearly every other field of human knowledge, and was widely considered the greatest science popularizer of the 20th century. He held a doctorate in biochemistry from Columbia University and was a tireless public speaker who lectured at hundreds of universities and conferences, appeared regularly on television with Johnny Carson and David Letterman, and spoke frankly about the future of technology, energy, and human civilization. His lectures are studied in university courses on science communication; his predictions about computing and information technology, delivered decades early, have been revisited with new urgency. Many of his conference talks were captured on reel-to-reel or consumer cassette tape by the host institution, with quality varying considerably depending on microphone placement and storage conditions.
Diffio addresses the high-frequency loss and ambient tape hiss characteristic of lecture-hall cassette recordings, restoring the warmth and intelligibility of Asimov's voice across the full range of his delivery.
John L. Lewis: Labor Speech and Radio Broadcast
Before (Original) · After (Diffio Enhanced)
John L. Lewis (1880–1969) was president of the United Mine Workers of America for four decades and the founding force behind the Congress of Industrial Organizations (CIO), which he established in 1935 to organize industrial workers the AFL had refused to represent. He directed the sit-down strikes of 1936–37 that brought General Motors and United States Steel to the table, funded organizing drives that expanded union membership from 12% to 32% of the non-agricultural workforce, and delivered some of the most powerful labor speeches ever recorded, including his 1947 congressional testimony on the Centralia Mine Disaster, in which he declared: "If we must grind up human flesh and bones in the industrial machine that we call Modern America… those who consume the coal… we owe protection to those men first." His voice (deep, theatrical, and deliberately paced in the grand Victorian oratorical tradition) was broadcast on CBS and NBC throughout the 1930s and 1940s and archived on 16-inch lacquer transcription discs, many of which have been played repeatedly and carry significant surface degradation.
Diffio removes the crackle and surface noise from the lacquer disc recordings and lifts Lewis's voice above the noise floor, restoring the physical presence and moral authority that made his broadcasts a defining sound of American labor history.
How Diffio Restores Historical Audio
Diffio is an AI audio enhancement tool trained to separate speech from the noise patterns introduced by analog recording media: surface crackle from lacquer and acetate discs, oxide shedding from aged magnetic tape, the characteristic hiss of consumer cassette, and the ambient room noise picked up by early microphone systems.
The underlying approach differs from older noise-reduction methods, which typically required manually sampling a section of silence and subtracting that noise profile from the entire recording, a process that often removed tonal content along with the noise and introduced artificial artifacts. Diffio applies a model trained across thousands of hours of degraded and restored audio pairs, learning to distinguish speech characteristics from noise characteristics at a level that simpler filters cannot reach.
Diffio 3.5, the current flagship model, is designed for exactly the kind of material shown in the demos above: recordings where the signal-to-noise ratio is low, the original recording conditions were far from ideal, and the historical value of the content makes preservation quality non-negotiable. In independent benchmarking on a 100-clip dataset, Diffio 3.5 achieved 22.5% more average MOS (Mean Opinion Score) improvement than Adobe Podcast, a meaningful gap when working with recordings where every decibel of clarity matters.
A few things Diffio does not claim: it cannot reconstruct audio content that was never captured. If a word was clipped in the original transmission, it remains clipped. If the original microphone was too far from the speaker, Diffio improves what exists; it does not add what is missing. The goal is fidelity to the original voice, not synthesis.
See how Diffio compares to Adobe Podcast in our Adobe Podcast alternative guide.
Who Uses AI Audio Restoration
The use cases for historical audio restoration extend well beyond professional archive institutions.
Archivists and libraries. Universities, government archives, public broadcasting organizations, and historical societies hold collections of deteriorated recordings that require restoration before they can be made publicly accessible. Digitization alone does not solve the noise problem; it preserves the degraded audio digitally. Restoration is the step that makes the content actually usable.
Documentary filmmakers and journalists. Historical documentaries depend on archival audio. A crackle-filled clip that distracts the viewer from content is often cut entirely, even when the historical value is significant. AI restoration makes clips usable that would otherwise be left on the cutting room floor.
Genealogists and family historians. Home recordings on old cassette tapes, reel-to-reel recordings of family events, dictation belts, and even wax cylinder recordings of grandparents' voices present the same noise problems at a personal scale. Restoring a family voice recording is not an institutional project; it is a personal one.
Podcasters and educators using archival clips. Podcasts that incorporate historical audio (history shows, political commentary, biography series) need clips that listeners can actually follow. A noisy clip that requires a transcript defeats the purpose of audio-first storytelling.
Institutions with public-facing digital archives. Museums, presidential libraries, university special collections, and public radio archives increasingly offer online listening to their holdings. The quality of that audio directly affects whether the public engages with those collections or abandons them after ten seconds of crackle.
How to Restore Your Own Recordings
Restoring old audio with Diffio requires no technical background, no software installation, and no specialized hardware.
Step 1: Upload your file. Go to diffio.ai and upload your audio or video file. MP3, WAV, MP4, and other common formats are supported. There is no limit on file age or format complexity.
Step 2: Select the Diffio 3.5 model. Choose Diffio 3.5 for maximum quality restoration. This is the model demonstrated in the historical recordings above. For faster processing at slightly reduced quality, Diffio 2.0 is also available.
Step 3: Download your restored file. Diffio processes the file and returns a cleaned version. Download the enhanced audio and compare it to your original. If you are integrating Diffio into an archival workflow, the REST API and Python and Node.js SDKs support batch processing at scale.
The free tier is available immediately: no credit card required, no sales call, no waitlist.
Start Restoring Your Recordings
The recordings above are not demonstrations of what AI audio restoration might eventually do. They are what Diffio does today on real historical material, using the same tool available to anyone who signs up.
Whether you are working with a century-old broadcast, a deteriorating cassette from a family archive, or interview recordings captured in a noisy room, Diffio applies the same model that achieved 22.5% more MOS improvement than Adobe Podcast, and returns a result in seconds.
No credit card required. Process your first files free. Batch processing via REST API, Python SDK, or Node.js SDK.
Sources
Library of Congress (Saving the Sounds of Radio); Harry S. Truman Library; Smithsonian Magazine (1948 Convention); National Air and Space Museum (Earhart Project Recordings); Voices of Democracy (John L. Lewis); U.S. Naval Institute (Earhart Radio); NPR Radio Diaries (Barkley).