Industries / Media, arts and recreation / Media and entertainment / Podcasting
Recording
Capturing the voices of a show, in one room or across a distance, and now often on camera as well.
7
Milestones achieved, from 1983 to 2025.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1983
Radio's phone-sync, written down
National Public Radio's audio engineer Skip Pizzi described the "phone-sync" in db magazine: the interview runs over an ordinary phone call, someone at the guest's end tapes only the guest on a good recorder, and the studio lines the two tapes up afterwards, using the phone track as a guide, so no phone-quality sound goes on air. The method is now called a tape sync or double-ender, and in 2014 a producer at the podcast 99% Invisible wrote that the show leans heavily on it.
LessonPizzi warned that the two tapes drift apart, "which they will always do," so any recording made on two separate machines has to be lined up again.
Credit
described by Skip Pizzi, NPR; the sources name no single inventor
2009
Skype calls become the studio
By 2009 producers were interviewing distant guests over Skype, an internet calling program, and recording the call on the host's computer with add-ons such as Audio Hijack. Transom's Jeff Towne wrote that internet calls are compressed, squeezed to travel efficiently, and can gurgle or drop out; Jason Snell, editor of Six Colors, used Ecamm's Call Recorder for Skype for "a dozen years," and in 2017 Zencastr's founder said that with such tools the guest still "sounded as if they're being recorded through Skype."
LessonA way of working built on another company's call app can disappear: Ecamm stopped updating Call Recorder for Apple's M1 Macs in 2021 and now says it "does not work anymore," and Microsoft shut Skype down on 5 May 2025.
Credit
Skype (Skype Technologies, later Microsoft); Ecamm
2015
The browser records each guest
With Zencastr, a host sent a guest a web link; the guest's browser then recorded the guest's own microphone on their own computer and sent the file to the host's Dropbox. That automated the double-ender with nothing to install, and its builder said it became possible only when browsers gained the ability to record from microphones in "just the past couple of years." Transom later wrote that Riverside, Ringr and Zencastr all "record audio locally, and automatically send those audio files to the producer."
LessonThe guest's computer is now part of the studio, and a local file is safe only once it has fully uploaded; the founder described a bug that left some backup files "basically unusable" on slower computers, and Zencastr's help pages say "not all of the audio for a guest makes it to dropbox properly" every time.
Credit
Zencastr, built by Josh Neilson (the spelling in a 2017 interview transcript), with work begun in 2014
2018
A mixer built for podcasts
RØDE's RØDECaster Pro went on sale in December 2018. It put four microphone inputs, phone calls with automatic mix-minus and recording to a memory card into one desktop box, so a show could be recorded without a computer; mix-minus is a return feed that sends callers everything except their own voice, so they hear no echo. RØDE called it "the world's first fully integrated podcast production studio," and in 2022 Transom's Jeff Towne wrote that its release "marked a pivotal moment in audio production" and that similar units from Tascam and Zoom followed.
Credit
RØDE
2020
Remote recording adds video
Riverside came out of stealth in November 2020. It recorded lossless audio and 4K video of each person on their own device, not over the call, so quality did not depend on internet speed, and it billed itself as "the first video podcasting platform," its own claim. By 2022 its users included The New York Times, iHeartMedia and Marques Brownlee, according to the Israeli business paper Globes, and The Week, summarizing The Wall Street Journal's Ben Cohen, says the turn to video podcasts began in the pandemic, when hosts recorded Zoom conversations and audiences watched them.
LessonMaking remote recording easier can lower the bar for quality; in 2023 the producer Mitra Kaboli told Transom, "I feel like the pandemic has eroded a quality standard," and that she missed "the era of tape syncs.".
Credit
Nadav and Gideon Keyson, Riverside (founded 2019, per Globes)
2022
One-click AI speech cleanup
Adobe Podcast's free Enhance Speech web tool takes an uploaded recording and uses AI to separate the voice from background noise. Gigazine tested it in December 2022 and found it removed fan rumble and breath pops, and in a March 2023 comparison for the Reynolds Journalism Institute it beat Audo Studio, Descript and Auphonic at reducing room noise but made the voice much louder.
LessonAI cleanup can change what was said, so a good room and microphone still matter: in a 2026 test by researchers from Nanjing University, Cisco and Horizon Robotics, transcripts of echoey audio cleaned by Enhance Speech V2 differed from transcripts of the clean original by 18.87% of words, against 4.55% for audio without the echo.
Credit
Adobe Research (Jiaqi Su, Zeyu Jin) and Adobe Podcast
2025
The double-ender moves into iPadOS
Apple's iPadOS 26 added Local Capture: during a call in any video-conferencing app, the iPad records its own user as lossless FLAC audio and HEVC video, while echo cancellation keeps the other callers out of the recording. Apple's guide for podcasters sums it up: "Each person records their own audio locally, then sends the file to be synced and edited later." Before this only one app at a time could use the iPad's microphone, so a call app and a recorder could not run together; Jason Snell recorded a full episode on iPads with the first test version and reported that it worked flawlessly.
Credit
Apple
Watch for
Studio-quality vocals from AirPods
Apple says a free firmware update, due with iOS 26, lets AirPods 4 and AirPods Pro 2 record "studio-quality" vocals for "interviewers, podcasters, singers, and other creators," crediting Voice Isolation, beamforming microphones, which listen in one direction, and its H2 chip.
Achieved whenan independent blind listening test, such as crowd ratings run under ITU-T P.808, the standard for rating speech quality online, rates AirPods recordings no worse than a wired podcast microphone recorded at the same time in the same room.
Claimed by
Apple, June 2025
Camera switching without an operator
Datavideo says its VoiceTRX50 box reads the audio from a podcast mixer and acts as a "Virtual Director," cutting to whoever is talking and to wide shots, and that it "eliminates the need for a dedicated camera operator." Switching cameras by voice is not new in itself: RØDE's RØDECaster Video had an automatic mode in 2024 that a Digital Camera World reviewer called "surprisingly good."
Achieved whenan independent review or a working show publishes full-length episodes cut live by such a device, with a count of the cuts an editor had to fix afterwards.
Claimed by
Datavideo, January 2026
Open
01
Blind test of speech cleaners
A public test of AI speech cleaners on real, deliberately bad podcast recordings. Record about 30 clips of your own voice or of consenting volunteers: a laptop microphone, an echoing kitchen, a phone on speaker, street noise, two people talking over each other. Run every clip through five or more tools, have strangers rate each result without knowing which tool made it, and separately count every word each tool changed.
The briefCredit · unclaimed02
Stress-test remote recorders
A repeatable rig that plays a known recorded voice into a "guest" laptop on each remote-recording service and injects network faults during each session: packet loss (missing pieces of data), a 30-second outage, a slow old laptop, and the guest closing the browser tab just after the session ends. Then measure how much of the guest's local track arrives intact, how many seconds are missing and how far it drifts out of sync.
The briefCredit · unclaimed03
Open set of recording faults
A Creative Commons collection of real two-person remote conversations by consenting volunteers, each with both local tracks, the call audio and a timed label file marking known faults: the wrong microphone chosen, speakers instead of headphones (so the other voice bleeds in), a Bluetooth headset, clipping (sound too loud for the recorder), an echoing room, a Wi-Fi dropout and an early tab close.
The briefCredit · unclaimed
Open 01 · the brief
Blind test of speech cleaners
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI speech cleaners on real, deliberately bad podcast recordings. Record about 30 clips of your own voice or of consenting volunteers: a laptop microphone, an echoing kitchen, a phone on speaker, street noise, two people talking over each other. Run every clip through five or more tools, have strangers rate each result without knowing which tool made it, and separately count every word each tool changed.
Why it is still open
The tests so far are small or synthetic. Production Expert (2023) used four clips with reader votes; the Reynolds Journalism Institute test (2023) relied on one author's ears; the 2026 StuPASE paper included one commercial tool, Adobe's, tested on simulated noisy mixtures with 70 samples in its listening test. The organizers of the URGENT 2025 speech-enhancement challenge say listener scores should be paired with checks for changed speech sounds to catch "hallucinations," made-up output. This research found no crowd-rated test of several commercial cleaners on real bad recordings that also checks for changed words.
Built with what exists
Adobe Enhance Speech, Descript, Auphonic, Audo Studio and iZotope RX as the tools under test; Microsoft's open-source P.808 Toolkit, with listeners recruited on Prolific or Amazon Mechanical Turk; Whisper Large v3 or ElevenLabs Scribe v3 (see the Note "Fewer word errors") to transcribe before and after. Check each tool's terms on publishing test results.
First step
Record 10 bad clips of your own voice in five rooms, run them through two tools, and pilot a 20-listener P.808 test.
Done looks like
30 clips through 5 tools, each result rated by at least 10 listeners under P.808, with average scores and 95% confidence ranges, a list of every changed word with its time, and all audio released under CC BY.
Open 02 · the brief
Stress-test remote recorders
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable rig that plays a known recorded voice into a "guest" laptop on each remote-recording service and injects network faults during each session: packet loss (missing pieces of data), a 30-second outage, a slow old laptop, and the guest closing the browser tab just after the session ends. Then measure how much of the guest's local track arrives intact, how many seconds are missing and how far it drifts out of sync.
Why it is still open
Makers and producers describe failures but publish no data. Zencastr's help page says guest audio does not always fully upload, and the production firm Resonate Recordings says it has lost sessions to "browser crashes" and "silent file-sync failures" but shows no test data for its ranking of services. This research found only user ratings on G2 and one anecdotal comparison of four sessions, and no independent controlled test.
Built with what exists
Riverside, Zencastr and Apple Local Capture, with a manual double-ender as the control; Linux tc netem or Clumsy on Windows to inject faults; the DAPS set of clean studio speech from Adobe Research as the reference voice (licensed CC BY-NC, so non-commercial use only); Audacity to measure gaps and drift. Check each service's terms before automating sessions.
First step
Run five sessions on one service with the Wi-Fi cut for 30 seconds at minute 10, and log what arrives.
Done looks like
3 services, 5 faults and 4 runs each, 60 sessions in all, with a published table of guest tracks that arrived complete, partial or lost, the seconds missing and the drift in milliseconds per hour.
Open 03 · the brief
Open set of recording faults
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A Creative Commons collection of real two-person remote conversations by consenting volunteers, each with both local tracks, the call audio and a timed label file marking known faults: the wrong microphone chosen, speakers instead of headphones (so the other voice bleeds in), a Bluetooth headset, clipping (sound too loud for the recorder), an echoing room, a Wi-Fi dropout and an early tab close.
Why it is still open
The closest public set, Adobe Research's DAPS, plays clean studio speech through a loudspeaker onto a tablet and a phone in seven rooms; it has no live calls and no second speaker, and its licence is non-commercial. New research still leans on simulated data: StuPASE's main test set mixes LibriSpeech readings with added noise and room echo. This research found no open set of real remote interviews with labelled faults.
Built with what exists
Apple Local Capture or any local recorder plus a free call app; Audacity for labels; whisper.cpp for transcripts (see the Note "A first stable number for local AI"); Zenodo for hosting.
First step
Draft the consent form and the fault checklist, then record five pilot pairs.
Done looks like
50 conversations of 10 minutes each, released under CC BY 4.0 on Zenodo with signed consent forms and a datasheet, in which each of the 7 fault types appears at least 20 times. The set doubles as test material for the speech-cleaner test.