Start with the clearest speech
Pick 10-30 seconds where one person is speaking normally. A short, clean section is usually more useful than a full file with music, edits, silence, or multiple speakers.
Free AI voice screening
Upload a voice note, record a quick sample, or check a short clip before you trust it. The detector gives a plain AI-risk signal for generated, cloned, or deepfake-style speech, then tells you when the result needs more context.
Check audio nowChoose or record audio to start.
This free AI voice detector is built for the moment when a clip sounds wrong and you need a quick second read. It checks a short speech sample and returns a probability-style result, not a courtroom answer. Use it for first-pass screening, then confirm important requests, identities, or claims through a channel you already trust. Live detection is powered by Modulate's Velma-2 Synthetic Voice Detection API.
Pick 10-30 seconds where one person is speaking normally. A short, clean section is usually more useful than a full file with music, edits, silence, or multiple speakers.
The server enforces the selected analysis window, then sends that short sample to Modulate's Velma-2 Synthetic Voice Detection API. Long uploads are trimmed for detection instead of being sent in full to the model.
Results appear as likely AI, likely human, or unclear. The confidence score is a screening signal; the source, speaker behavior, and recording quality still matter.
Voice scams and synthetic narration often show up in short clips: a voicemail from an unknown number, a social post with no source, a polished ad voiceover, or a message that asks you to act fast. Those are good moments to run a check. The goal is not to accuse anyone; it is to slow the decision down and look for more evidence.
Use the detector when a voice note asks for money, credentials, secrecy, urgent action, or a change in normal behavior. A likely AI result should trigger extra verification through a trusted channel.
If you review ads, video narration, podcasts, or social clips, a short speech sample can help flag audio that may have been generated or heavily processed by AI voice tools.
Before reposting a clip that claims to capture a real person, check the voice signal and review the source, platform, timestamp, and whether the message can be confirmed elsewhere.
Treat the result as a risk signal. "Likely AI" means the analyzed speech has patterns that resemble synthetic or generated voice. "Likely human" means the selected window did not show a strong AI signal. "Unclear" means the detector could not make a useful call, which often happens with short clips, compression, background music, speaker overlap, or aggressive noise reduction.
A single speaker, low background noise, and a natural 10-30 second segment usually produce a more useful detection signal than music-heavy, clipped, or heavily compressed audio.
Do not rely on one score for employment, legal, financial, law enforcement, or safety decisions. Use the detector alongside provenance checks and direct confirmation.
If the result is unclear, try a cleaner section of the same voice. Avoid clips with overlapping speakers, noise suppression artifacts, phone hold music, or loud background sound.
The browser prepares a short WAV sample before upload, and the server enforces the selected analysis window again before sending it to Modulate's Velma-2 Synthetic Voice Detection API. The app is designed for short screening, not long-term audio storage. Do not upload private, confidential, or legally sensitive audio unless you are comfortable having a short sample processed by Modulate. If the result panel identifies the provider as "Mock," the app is operating in demo mode and the result was not produced by Velma-2.
A detector can miss a well-made clone, and it can also flag real speech when the recording is poor. For anything that affects money, access, employment, reporting, legal claims, or personal safety, use the result as one data point. Call the person back on a known number, check the original source, ask for a live response, or compare the clip with trusted recordings.
AI Voice Detector is free-first. Basic checks remain available without checkout: 3 checks per day without login, or 10 free checks per day with optional Google login. Pro is $9.90 per month and includes up to 20 checks per day for signed-in users. If you only need to check the occasional voice note, the free limit is meant to be enough; Pro is for repeated review work.
No. It can point to AI-like speech patterns, but it cannot prove who made the clip or whether a recording is authentic. Use it with source checks and human review.
No. You can run basic checks without an account. Signing in with Google only raises the free daily limit.
Use a speech-only section with one speaker. Avoid songs, crowd noise, phone hold music, and clips where the suspicious voice is only a small part of the file.
For account, billing, subscription, refund, privacy, or detection questions, email bingkun.zhao@gmail.com. We aim to respond within three business days.