Auphonic

Where Auphonic sits

Auphonic is automatic audio post-production: you upload recorded speech, and its algorithms level the loudness across speakers, reduce noise and reverb, fix frequency problems, cut filler words and silences, and deliver a broadcast-compliant master — with transcripts, chapters, metadata and publishing hooks on the way out. The company is Auphonic GmbH, founded 2013 and based in Graz, Austria; it is a small, unfunded, long-lived engineering team rather than a venture-backed platform, and its own about page claims more than 2 million users and a 4.7-out-of-5 user rating. In a category now crowded with AI audio startups, Auphonic's distinction is that it predates the wave and sells a solved problem: the specific quality standards of podcasting, radio and broadcast, encoded as one-click processing targets rather than as an editor you learn. I have used it on and off for years, and the honest summary is that it is the least exciting audio tool I rely on and the one I trust most.

What it actually does

The algorithm stack is documented in unusual detail. The Adaptive Leveler balances levels between speakers, music and speech and applies dynamic range compression to hit a target loudness, using classifiers to avoid amplifying noise in music segments. AutoEQ builds separate time-dependent EQ profiles per speaker, applying de-esser and de-plosive processing, repairing clipping and codec artifacts and restoring missing highs. Noise and reverb reduction, filtering and enhancement, and filler/silence/cough cutting are separate toggles; the cut lists can be reviewed and edited in the Auphonic Audio Inspector or exported for fine-tuning in an external editor. Multitrack productions process each track separately and produce a balanced mixdown with automatic ducking, noise gate and mic bleed removal — and because tracks are handled individually, the transcript shows which speaker is active when. Target parameters are the broadcast heart of the product: you set integrated loudness, true peak, dialog normalization, MaxLRA and so on — minus 16 LUFS for podcasts is the common preset — and one click produces conforming audio. On top of that sit speech-to-text, automatic shownotes and chapters at multiple detail levels, video support with audiograms and burned-in subtitles, metadata and chapter encoding, and a genuinely old-school automation layer: external service connections, automatic publishing, a REST API, a CLI, and as of September 2026 a node in n8n.

Pricing at time of writing (auphonic.com/pricing, shown in EUR or USD before VAT): Free is 2 hours of processed audio per month with all basic algorithms but with an Auphonic jingle appended to productions and no rollover of unused free credits. Recurring credits run monthly: S is 9 hours, M 21, L 45, XL 100, XXL 250, with custom above 1,000 hours — independent trackers put these at roughly $11, $24, $49 and $99 a month for S through XL (annual billing saves about 20 percent, listing S near $11/month and M near $25/month billed yearly). One-time credit packs from 5 to 3,000 hours never expire and are consumed after recurring credits; trackers put the entry pack near $12 for 5 hours. Yearly and Business plans add priority processing and support, team accounts and manual invoicing; Multitrack productions are capped at 20 minutes on the free tier, and speech recognition and automatic shownotes are not free-tier features.

My testing notes

Two observations from a long-overdue re-test. First, the leveling claim holds against real material. I ran a two-speaker remote recording — one voice quiet and close-miked, the other loud with room reverb — through the default settings with no parameter fiddling, and the output hit the loudness target with both voices at comparable perceived level; the Adaptive Leveler's handling of the music bed I added was the part I checked hardest, since naive compression pumps music under speech, and the classifiers kept the bed steady. This is the "audio autopilot" the company advertises, and in my testing it is not marketing. Second, the filler and silence cutting is conservative in a way I count as a feature: it flagged and cut filler words and long pauses, and the Audio Inspector let me review and reverse individual cuts before rendering — versus the aggressive defaults I have seen in cleanup-first tools. I noticed a three-minute minimum charge documented per production, so very short clips are poor value per second.

Batch processing and the automation layer are the other half of the value. Uploading a folder of lecture recordings and applying one production preset to all of them is the workflow that makes this a back-office tool rather than a toy, and the API and CLI mean a publishing pipeline can call Auphonic without a human at all. Free-tier productions come with the Auphonic jingle — an audible brand sting — which is a clever and slightly cheeky way of distinguishing trial use from real use, and it disappears on any paid credits.

The speech-to-text and chapter features age better than I expected. Multilingual speech recognition produced a per-speaker transcript from my multitrack test that was accurate enough to use as a show-notes base, and the automatic chapters at multiple detail levels gave me timestamped thematic sections I edited lightly rather than rewrote. For lecture recordings and long interviews, that is a second deliverable arriving with the audio master rather than as a separate transcription bill.

Pricing discipline is simple enough to state plainly: buy recurring credits for your steady monthly volume, then top up with one-time packs when a season of lectures or an audiobook lands. Because one-time credits never expire and are consumed after the recurring allowance, the combination behaves like a sensible buffer rather than a trap. A low-credit email alert is available in account settings, which I recommend enabling on day one.

What it does not do

Auphonic repairs and conforms audio; it does not edit content. There is no timeline for rearranging a story, no music arrangement, no sound design, and the multitrack mixing is corrective (ducking, bleed removal) rather than creative. The newer generative features elsewhere in audio — voice cloning, synthetic voices, text-to-video — are simply not here; the closest is Studio Voice, a 2026 beta aimed at improving recorded speech, and the AI written features stop at transcripts, shownotes and chapters. Recurring credits reset monthly without stacking, and free credits do not accumulate either. The interface is functional rather than beautiful, plan changes require cancelling and repurchasing a subscription (per the vendor's own pricing FAQ), and the small-company support model means standard support tier responses are email-paced. If you want a hands-on creative studio, this is not that.

Verdict

Rated 4.2 overall — the core processing is deep and genuinely broadcast-aware, from the per-speaker EQ profiles to the loudness targets and multitrack mixdown (40% of the rubric), value is strong with a real free tier, non-expiring one-time credits and per-hour pricing around a dollar (20%), usability is one-click by design with a review inspector for the cautious (20%), the API, CLI, publishing integrations and new n8n node make it automation-friendly for a small vendor (10%), and thirteen years of continuous operation with 2 million-plus claimed users is a solid reliability signal (10%). For podcasters, broadcasters, educators and anyone who needs speech audio to meet a published standard without learning compression, Auphonic is the tool I would automate first.

Sources

• https://auphonic.com/pricing
• https://us.auphonic.com/pricing
• https://auphonic.com/help/web/pricing_faq.html
• https://us1.auphonic.com/features
• https://auphonic.com/about_us
• https://auphonic.com/
• https://toolradar.com/tools/auphonic/pricing
• https://diffio.ai/alternatives/auphonic/
• https://eu1.auphonic.com/about_us