Cleanvoice AI
AI podcast editor that strips background noise, filler words, mouth sounds and dead air from audio and video recordings, then transcribes and summarises them. Available as a web app and an API, with EU hosting.
What is Cleanvoice AI?
Cleanvoice AI is an automated post-production service for recorded speech, run by Sigmoid Creativity S.R.L. from Bucharest and online since 2021. Its pitch is blunt: what takes four hours of manual editing should take about ten minutes, and you should be able to be a podcaster rather than an editor. You upload an audio or video file, the system analyses it, and you download a cleaned version.
The cleanup covers what usually forces an editor to stop every few seconds. Background noise — traffic, dogs, doors, wind, hiss, hum, mic buzz — is removed, along with echo, reverb and distortion. Filler words, hesitations, stutters, mouth sounds and breaths are cut, with silence put back in their place so the delivery still sounds natural. Long silences and dead air go too. Studio Sound then balances levels and evens out the tone. Formats accepted include .wav, .mp3, .m4a, .flac and .mp4, up to 1.5 GB per file, or 2 GB across a multitrack upload.
Unlike editors built around a transcript, Cleanvoice does not ask you to edit text to edit audio: you upload the file and transcription happens on its own. That transcript feeds the post-production side — summaries, show notes, chapter markers and social content — so an episode can be finished without switching tools. Guests recorded on separate tracks are handled together and resynchronised into one episode. Editors who prefer to keep control can export markers or timelines into Adobe Audition, Adobe Premiere, DaVinci Resolve, Reaper, Audacity or anything that reads EDL files, and a Mute Edits template silences passages instead of cutting them, which keeps video in sync.
There is a genuine developer side: a REST API with JavaScript and Python SDKs, an interactive playground, asynchronous jobs with automatic polling, webhooks and delivery to your own S3 storage. The company claims more than 30 brands on the API and over 15,000 podcasters overall. A Make connector covers no-code workflows. Around it sit free utilities — podcast name and question generators, an audio trimmer, joiner, mic test and audiogram maker. Security is a stated selling point: EU hosting, ISO 27001, a public DPA, a 99.5% uptime SLA, seven-day retention and no training on customer data.
What it does
- Remove background noise, hum, hiss, wind, echo, reverb and distortion from a recording
- Cut filler words, hesitations, stutters, mouth sounds, breaths and long silences
- Apply Studio Sound to balance levels and give a recording a studio-grade tone
- Transcribe audio or video automatically and export the text in several formats
- Generate an episode summary, show notes, chapter markers and social posts
- Edit several guest tracks at once and keep them in sync as one episode
- Export markers or timelines to Audition, Premiere, DaVinci Resolve, Reaper or Audacity
When to use Cleanvoice AI / When not to
A quick filter to help you decide if Cleanvoice AI is the right fit.
When to use Cleanvoice AI
- Solo podcasters who record and edit their own episodes and want the cleanup done automatically
- Media agencies and video editors handling recurring volumes of interviews and video podcasts
- Product and engineering teams embedding audio cleanup in their own apps through the API and SDKs
- Audiobook narrators and voice-over artists working outside a treated studio
- Researchers, journalists and teams recording remote interviews or Zoom calls with uneven sound
When not to use Cleanvoice AI
- Anyone looking for cloud storage or an archive: files are deleted after seven days by design
- Users who need synthetic voices or voice cloning, which the tool does not do at all
- Speakers working in languages beyond the five confirmed ones, where results are not guaranteed
- People under 18, who are barred from the service by the terms and conditions
- Buyers who need a permanent free tier rather than a 30-minute trial followed by paid credits
How to use Cleanvoice AI
A typical end-to-end flow, from setup to results.
- Open the app and drag your audio or video file in — no account or card is needed to try it
- Pick a ready-made template, or select exactly which treatments you want applied
- Choose whether unwanted passages are cut out or simply muted, to keep video in sync
- Upload several episodes at once if you are working through a backlog
- Let the job run: processing is asynchronous, so you do not have to sit and wait
- Download the cleaned file, or export markers and timelines into your own editor
- Optionally generate the transcript, summary, show notes and chapter markers
- For automation, create a developer account and copy your API key from the dashboard
- Install the JavaScript or Python SDK, or call the REST endpoint directly, and try it in the playground
- Collect results by polling, which the SDK handles for you, or through a webhook to your storage
Pros & Cons
Pros
- You can test it on a real file without signing up or entering a card, then get 30 free minutes on registration
- No transcript-to-edit workflow: you upload the file and the transcription happens by itself
- Every paid plan includes every feature, so pricing buys volume rather than functionality
- Serious data posture for the category: EU hosting, ISO 27001, a public DPA, named subprocessors and no training on customer data
- Pay-as-you-go credits stay valid for two years, and unused subscription credits roll over up to three times the plan limit
- A real developer offering with SDKs, a playground, webhooks and a contractual 99.5% uptime SLA
- Fine-grained control through templates, per-feature selection, mute-instead-of-cut and exports to professional editors
Cons
- No permanent free tier: after the 30-minute trial, everything is paid
- All fees, subscriptions and credit purchases are final and non-refundable
- The confirmed language list for filler word removal is five languages, against the 20+ advertised on the homepage
- The privacy policy sits on a third-party JavaScript portal and is awkward to read or archive
- No dedicated GDPR or DPO contact address is published anywhere on the site
- No mobile app and no browser extension: it is a web app and an API
- Files disappear after seven days, and guest uploads made without an account after only four hours
Pricing & Plans
There is no permanent free plan. A free trial covers 30 minutes of processed audio or video without a card, and a further 30 minutes of credits is granted on sign-up. Paid access starts at EUR 10 per month for the 10-hour subscription, or EUR 10 for a one-off pay-as-you-go pack of 5 hours whose credits remain valid for two years. All prices are quoted excluding VAT and every paid plan includes every feature.
- Free Trial — 30 minutes of processed audio
- basic features
- no credit card required
- EUR 0
- Pay as You Go 5 hours — EUR 10 / USD 11
- all features
- credits valid for two years
- Pay as You Go 10 hours — EUR 18 / USD 20
- all features
- Pay as You Go 30 hours — EUR 40 / USD 45
- all features
- Subscription 10 hours per month — EUR 10 / USD 11 monthly
- or EUR 100 / USD 110 yearly
- Subscription 30 hours per month — EUR 26 / USD 30 monthly
- or EUR 260 / USD 300 yearly
- Subscription 100 hours per month — EUR 85 / USD 90 monthly
- or EUR 850 / USD 900 yearly
- Custom Plan — from USD 0.20 per hour above 200 hours per month
- custom API endpoints
- priority support and tailored billing
- by appointment
Data, GDPR & hosting
A consolidated view of how Cleanvoice AI handles your data.
GDPR overview
GDPR compliance is claimed explicitly and backed by documents rather than badges. The homepage states compliance with the GDPR, data stored in the EU and ISO 27001 certification. A full Data Processing Agreement is published at cleanvoice.ai/dpa, drafted around Article 28(3): it defines Cleanvoice as processor or subprocessor, commits it to act only on the client's documented instructions, and undertakes to assist with data subject rights requests. Annex 2 lists the technical measures — pseudonymisation, encryption, access control, event logging, restoration testing — and Annex 3 names all six subprocessors with their data locations. The agreement is governed by Romanian law with exclusive jurisdiction in Bucharest. The operator is established in Romania, so no Article 27 representative applies. No dedicated DPO or privacy email is published; the privacy policy sits on an external Privasee portal.
Who owns the data?
You keep everything you bring and everything the service produces. Section 7 of the terms states that you retain all right, title and interest, including intellectual property rights, in your Content, and that the same applies to the Results generated from it, with Cleanvoice explicitly claiming no ownership over them. What the company receives is a worldwide, non-exclusive, royalty-free licence to host, copy, process and transmit that material, strictly limited to delivering the service you asked for. Subprocessors may handle the files only to the extent needed. Aggregated, anonymised and de-identified statistics fall outside this and remain usable by the company.
Reuse rights
Yes. Because you keep the rights to both your uploads and the cleaned output, you can publish, sell, license or reuse the results commercially without asking permission, and the terms confirm that Cleanvoice claims no ownership over them. The company's own use is narrowly bounded: your content serves only to deliver the service you requested, except where the law or a binding order from a competent authority requires otherwise. Crucially for a speech tool, the developer pages state that customer files are never used to train models, the company having recorded its own dataset in professional studios, and section 11.7 of the terms repeats that no retained content feeds AI training. Named subprocessors include DigitalOcean, Hetzner, Cloudflare, Bunny, RunPod and Anthropic Ireland.
Data retention & training
Hosting summary
Cleanvoice states that data is stored in the EU, and the developer pages specify that both servers and GPUs are located there. The published DPA names every subprocessor and where each one holds data. DigitalOcean LLC and Hetzner Online GmbH provide infrastructure and storage with data located in Germany. Cloudflare Inc. handles storage, content delivery and DDoS protection, Bunny (BunnyWay d.o.o., Slovenia) provides storage and delivery, RunPod Inc. supplies cloud compute, and Anthropic Ireland Limited performs AI processing on user queries and results — all four with data located in the European Economic Area. The controlling entity is Romanian and the DPA is governed by Romanian law, with exclusive jurisdiction in Bucharest. On-premise deployment is offered on request. Note that the website's own IP resolves to an anycast CDN node announced from the United States, which reflects content delivery rather than where recordings are stored.
Things to keep in mind
Risks and trade-offs to weigh before adopting Cleanvoice AI.
- Automated cleanup can quietly reshape meaning: removing hesitations and pauses changes how a speaker comes across, which matters for interviews, journalism and anything evidential
- Leaning on the tool erodes editing skill over time, and the judgement about what should stay in a recording is exactly what an editor is for
- Recordings usually contain other people's voices, and the terms make you responsible for having the legal grounds and consents to process their personal data
- Files vanish after seven days, four hours for guest uploads, so treating the service as storage will lose work
- Payments and credits are non-refundable, and unused subscription credits are lost at the end of the billing period after cancellation
- The advertised 20+ languages is not backed by the site's own confirmed list of five, so results outside them are unpredictable
- The ISO 27001 claim is stated without a certificate number or certifying body, and the privacy policy lives on an external portal that is hard to read or archive
Setup & Integrations
Technical difficulty
Very low for the web app: you drag a file in, pick a template or tick the treatments you want, and download the result. No account is needed to try it and the site claims no editing jargon to learn. The API is aimed at developers but stays light — the company advertises setup in five clicks with copy-paste code, the API key is configured for you, and JavaScript and Python SDKs or a plain REST call all work. Jobs run asynchronously, with polling handled by the SDK or results pushed to a webhook. A playground lets you test endpoints before writing any integration.
Deployment
Integrations
Supported languages
Behind Cleanvoice AI
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Cleanvoice AI.
Frequently asked questions
Can I try Cleanvoice AI before paying?
How long are my files kept?
Are my recordings used to train the AI?
Which languages are supported?
Which file formats and sizes can I upload?
Do I have to transcribe before editing?
Can I keep control over the edits?
Is there an API, and does it have usage limits?
Can I get a refund if I stop using it?
Is there a service level commitment?
Should you pick Cleanvoice AI?
Cleanvoice AI is a specialist rather than a generalist, and that is its strength. It does one thing — turning a raw speech recording into something publishable — and it covers that job from noise removal through to show notes without asking you to leave for another tool. The decision not to build around a transcript is the clearest point of difference from most AI podcast editors: you upload a file rather than edit text, which suits people who think in audio rather than in documents.
The data posture is unusually legible for this category. EU hosting, a published DPA that names all six subprocessors and their data locations, seven-day retention with an optional zero-retention mode, an explicit statement that customer files never train the models, and a contractual uptime commitment together answer the questions a procurement team actually asks. Ownership is equally clear: you keep the rights to both your uploads and the cleaned output.
Two reservations deserve weight. The advertised reach of more than 20 languages is not matched by the five the site actually confirms, and anyone working outside English, French, German, Romanian or Arabic should test before committing. And the privacy policy is delegated to a third-party JavaScript portal with no dedicated GDPR contact, which sits oddly against otherwise careful legal documentation. Commercially, the credit model is straightforward but unforgiving: nothing is refundable, and unused subscription credits expire when you cancel.
For a solo podcaster, a media agency or a team that needs audio cleanup inside its own product, the free trial costs nothing to run on a real file, and that is the sensible way to judge it.
- Choosing a selection results in a full page refresh.
- Opens in a new window.