Most comparisons of voice detectors rank them by accuracy figures the vendors publish about themselves. We have not tested competing products, so we do not rank them. What we can do is set out, from each vendor's own public pages, which job each kind of tool is built for — because a detector built for a different job is the most common reason people get a misleading answer.
| Type | What it answers | Examples (vendor pages) | Where it falls short |
|---|---|---|---|
| Generator-specific classifier | Was this made with our product? | ElevenLabs AI Speech Classifier — built to detect ElevenLabs-generated speech; ElevenLabs says it cannot detect other providers' audio and does not reliably classify its v3 model. Its newer Audio Detector checks for a watermark first. | A “not ElevenLabs” result says nothing about every other voice tool. Watermarks only exist on audio generated after they were introduced. |
| Live-call and meeting detection | Is the caller or participant on this live call synthetic? | Pindrop Pulse (contact centres, and meetings on Zoom, Teams and others); Hiya AI Voice Detection (phone calls, via apps and carriers). | Built into call infrastructure and sold to businesses. Not a way for an individual to check a file they were sent. |
| Browser extension | Is the voice playing in this browser tab synthetic? | Hiya Deepfake Voice Detector (Chrome); NordLabs AI Voice Detector (Chrome, analyses tab audio locally). | Works on audio you can play in a tab. A voice note on your phone or a file on your disk needs another route. |
| File upload and API | Is the speech in this recording synthetic? | Resemble Detect (API and dashboard; Resemble also sells voice generation); Reality Defender (API and web app, audio among other media); TextSight. | Judges a recording after the fact. Cannot warn you during a call. Results on compressed audio such as voice notes are weaker for every tool in this group. |
Every line here comes from our product code, not marketing.
| TextSight AI Voice Detector | |
|---|---|
| Type | File upload, plus an API on Amplify and above |
| What it answers | Is the speech in this file synthetic? Returns a verdict and a 0–100 AI-likelihood score |
| Generators | Not tied to one vendor. It does not name the tool, and we have not published a per-generator detection rate |
| Input | MP3, WAV, M4A, OGG or FLAC, up to 10 MB. No video files, no links, no live audio |
| How much is analysed | The first 60 seconds of speech in each clip, stated on every result |
| Live calls / extension | No and no |
| Audio retention | Processed and discarded, not stored, not used for training |
| Access | First check free with no signup, 3 a day with a free account, public self-serve plans from $19 a month — see pricing |
| Evaluation | No published accuracy figure yet. Compressed audio can be wrongly flagged — see the accuracy page |
Whatever you choose, the questions in how to choose a voice detector apply: ask any vendor how their accuracy was measured, on what audio, and what happens with compressed or phone-quality recordings.
The questions to ask any vendor, and a ten-minute test you can run yourself.
Read the guide →What moves a result up or down, and what false positives look like.
See our position →The full routine, from best copy to verifying the person.
Read the steps →Your first check is free with no signup — 3 a day with a free account. Your audio is processed and discarded, never stored.