Frequent Questions & Answers

Yes, Addavox prioritizes your security. All videos and movies are encrypted both in transit and at rest.

No. We never sell your content, and we never share your videos or your finished localized work with third parties. Your files are kept in our own encrypted storage, hosted on AWS. To run each step of a job, we send only the audio or text that step needs, such as the speech audio for transcription or the script for translation, to the AI providers we use: AWS, Google, Microsoft Azure, Deepgram and Anthropic. They process it on our behalf and never receive your whole video or your final work. We only collect the personal information we need to provide the service, such as your name and email address, and card payments are handled by Stripe, so we never store your card details.

Yes — up to 3 hours and 12 GB per file. Professionals localize full-length features within that, and anything larger can be resized or split into multiple files. If a file does exceed the limits it is rejected before processing and you are refunded automatically, so you are never charged for work that could not run. Longer videos take more time through the AI, but the time saved against manual localization is still significant.

We transcribe in 80 source languages across 154 regional variants, and produce voice in 80 target languages across 154 regional variants, with 2,782 voices available. New languages are added regularly.

Yes, Addavox automatically creates a project when videos are uploaded. You can track progress and add and revoke access for reviewers.

Yes! You can assign professional reviewers to specific projects and languages within the app and monitor their progress live.

Addavox voice matching is built for localization, which makes it a poor tool for deepfakes by design. It only creates a voice in a language different from the one the speaker is speaking, so it cannot regenerate someone's voice in their own language. Voice-matched localization requires a recorded consent attestation before processing begins. Voices are matched in memory, one sentence at a time, and no voiceprints, voice models or cloned voices are stored for reuse. Our EULA prohibits deepfakes and impersonation.

Transcripts, localized subtitles, voiceover tracks for each language as well as a mixed track that combines the background music, effects, and the target language voiceover.

Currently, Addavox uses 10 AI models, with more in development. Our goal is to make audio and video localization as efficient as possible. We only develop AI tools that can deliver faster, higher-quality results or improve upon what can be done manually.

Yes, but we advise against it. In our own testing, transcription and machine translation are 95% to 98% accurate, but errors can compound: a transcription mistake can lead to an inaccurate translation. For the best results, have a professional review and refine the AI-generated output.

Voice matching captures each speaker's tone, vocal qualities and emotion sentence by sentence, so a whisper is matched as a whisper and a shout as a shout. It then creates a new voice in the target language that sounds native to speakers of that language. You can adjust some aspects of tone in the editor, and you can bring in your own voice artists for any line that needs a human performance.

Voice matching keeps each speaker recognizable, with their tone, vocal character and emotion, while making them sound native in the target language. It is deliberately not an exact copy of the original voice: that is what voice cloning does, and it is what makes cloning useful for deepfakes. There is no minimum length: Addavox can voice-match a single spoken word, and quality improves with longer utterances of about 3–5 seconds or more.