Reply to thread

Message: [QUOTE="Bombastus, post: 91916, member: 2178"] TonalBrain says its plugin keeps raw audio on your computer while sending spectral analysis data to its servers for mix calculations. No audio files are uploaded, according to the company, and the raw recording remains on the local machine. The distinction is useful because “local audio” and “fully offline” are not the same promise. The plugin first analyzes frequency and dynamic characteristics locally, then sends derived information out for the server-side calculation that returns mix settings. So [B][URL='https://goldmidi.com/community/threads/tonalbrain-mix-gains-tape-saturation-and-pro-tools-aax.77080/']TonalBrain’s local audio analysis model[/URL][/B] reduces exposure of the original recording without removing the network step entirely. Producers working with unreleased music should judge those two parts separately. [HEADING=2]Local analysis still creates a network boundary[/HEADING] A raw WAV, vocal stem, or multitrack file contains the original recorded signal. Spectral analysis data is a derived representation of characteristics inside that signal, and TonalBrain says only the data needed for its mixing decisions is sent to its servers. Its FAQ also says the company does not store or share audio content or analysis data beyond processing the request. Derived data deserves precise language, though. Spectral representations can preserve meaningful structure from audio, depending on how they are constructed and how much information they retain. [B]Spectral representations can retain reconstructable detail[/B] [URL='https://arxiv.org/abs/2010.05646?utm_source=chatgpt.com']HiFi-GAN paper on mel-spectrogram inversion[/URL], but TonalBrain does not publicly say it sends mel-spectrograms or provide enough detail to claim its own payload can be turned back into a song. So the sensible reading is narrower. TonalBrain states that it does not upload your audio files, while some derived spectral data crosses the network for processing. Calling the service fully local would go too far because the mix calculation depends on a remote server. [HEADING=2]The privacy policy covers more than spectral data[/HEADING] TonalBrain’s privacy policy lists additional plugin usage information beyond the analysis payload. It includes the plugin build, DAW and operating system versions, crash and error logs without audio content, analysis-request timestamps, instrument names or descriptions entered into the plugin, approximate country or region, and an anonymized device identifier. None of those fields is the same thing as a recorded stem, but they still describe parts of your production environment. Instrument names deserve attention because producers often put artist, song, or client identifiers directly into track labels. A field such as “Lead Vox Sarah Final” exposes more context than “Audio 17,” even though neither contains recorded audio. Account activity adds another layer. TonalBrain stores account information such as email, credit balance, purchase history, and device activation information, while payment handling goes through Stripe. The website uses self-hosted Umami analytics, and signed-in browsing sessions can carry an account identifier even though the company says it does not attach your email address or name to those analytics sessions. Several service providers also sit around the product rather than inside the audio path itself. TonalBrain names Supabase for database and authentication, Stripe for payments, and Cloudflare Turnstile for abuse prevention. Its policy also says personal information may be transferred to or processed in countries other than the country where you live. [HEADING=2]Your session rules decide whether the design is acceptable[/HEADING] For your own demos, the important boundary may simply be whether the company receives the actual recording. TonalBrain’s stated architecture clears that narrower test because it says raw audio remains on the machine. A producer who only wants to avoid uploading stems may therefore see the design very differently from someone who requires a completely disconnected workstation. Client sessions can impose tighter rules. An NDA, label security policy, film-production agreement, or internal studio rule may prohibit remote processing of session-derived information even when the original files never leave storage. In that environment, sending spectral analysis data to a server can matter because the contract may care about transmission, not just file uploads. TonalBrain also says users retain ownership of audio processed with the plugin and that it claims no rights in the music or recordings. Ownership and confidentiality are separate issues, however, so retaining copyright does not answer whether a particular client permits cloud-assisted processing. The safer operational test is the wording of the agreement governing the session. One gap remains in the public explanation. TonalBrain describes the transmitted material as spectral data needed for mixing decisions, but it does not publicly spell out the exact feature format, granularity, server location for each request, or technical measures that would let an outsider judge how informative the payload is. If your project requires proof that no session-derived data leaves the workstation, the current design does not meet that strict standard because remote calculation is part of the workflow. [/QUOTE]

Name