This policy covers one feature of the ViralDNA.io platform: ViralDNA SRT Maker, the tool at viraldna.io/srt-maker that turns a voiceover into subtitles (SRT, VTT or a timestamp list).
It has its own policy because it works differently from most tools on the platform: your audio file never leaves your computer. The speech is transcribed by a model that runs inside your browser, and only a small amount of text and numbers reaches our server to measure the timings. We store none of it.
Everything else about your ViralDNA account, including sign-in, billing and the other tools, is covered by the Platform Privacy Policy. This document adds to it and does not replace it.
The short version.
Your audio is never uploaded or stored. Your browser transcribes it on your own machine. To measure exactly when each sentence starts, it sends our server a loudness summary (one number per 10 milliseconds, from which the sound cannot be rebuilt), the transcribed words with their timings and, only if you pasted it, your script. The server returns corrected timings in the same response and keeps no copy.
When you choose an audio file, your browser decodes it and converts it to a 16 kHz mono signal in memory. Nothing is transmitted at this stage.
When you press the button, your browser runs the Whisper speech recognition model (an open model published by OpenAI, in the ONNX version published on Hugging Face) locally. The first time, your browser downloads the model files from Hugging Face and keeps them in its own cache, so later runs do not download them again. The transcription itself runs on your computer.
Speech models give approximate word timings. To make them exact, your browser then sends our server the data listed in section 2. Our server finds the silent moments in the loudness summary, matches each one to the word that follows it, and returns the corrected timings. Your browser builds the subtitle files from them and saves them straight to your computer.
Exactly the following, and nothing else, in one request each time you generate or re-apply a script:
| Field | Example | Why it is needed |
|---|---|---|
| Loudness summary | One byte per 10 ms; 60 minutes of audio = about 360 KB (about 480 KB as sent, base64) | To find where the silences are. It records only how loud each 10 ms was; the sound cannot be played or rebuilt from it. |
| Length of the audio | 5,154,816 samples | To check the summary is complete. |
| Transcribed words and their approximate timings | "Berlin" 0.20 → 0.62 | The words the model heard, so each silence can be matched to the word that follows it. |
| Your script, only if you pasted one | The text of your voiceover | So the subtitles use your exact words; it is matched to the transcribed words. |
| Spoken language code | tr | Word matching depends on the language (for example Turkish dotted and dotless i). |
Because the request is made while you are signed in, it also carries your session, which tells our server whether your account is entitled to use the feature. That check reads your plan; it writes nothing.
The transcribed words and your script are text from your voiceover. We do not read, index, profile or retain them, and they are not sent to any AI provider.
Nothing. The data described in section 2 exists only for the moment it takes to compute the timings. It is not written to a database, not written to disk, not placed in a queue, and not copied into an analytics system. There is nothing for you to request, export or delete afterwards.
Two ordinary exceptions apply, and they are technical rather than about your content:
If we ever change how this feature handles data, we will update this page and change the date at the top before the change takes effect. If a change ever meant that your audio would be transmitted or stored, we would say so in plain language on the feature page itself, not only here.
Questions about this policy: support@viraldna.io.