Timbre is Plurall AI's synthetic speech detection model. Upload an audio clip and it returns a calibrated likelihood that the speech was generated rather than recorded, alongside the error rates it was measured at.
Drop an audio file here
or
wav, flac, mp3, ogg
It is sent to our scoring service over an encrypted connection, decoded, scored, and deleted. Nothing is retained: the log records a content hash, the duration and the verdict, and never the audio or the filename. Audio shorter than one second, effectively silent, or below an 8 kHz source rate is refused rather than scored.
A detector is only as good as the conditions it was measured under, so the conditions are stated rather than summarised.
| Worst held-out generator conditionSix generator-disjoint and speaker-disjoint folds. The headline is the worst fold, not the mean of 1.46 %. | 3.16 % EER |
| Reserved non-vocoder conditionOne attack built without a neural vocoder, held out of training entirely and scored in every fold. One generator, and the same clips each time. | AUC 0.9999 |
| Held-out equal error rate of the shipped headOn the partition the operating points were derived from. | 0.293 % |
| Five Google Cloud TTS voice familiesFull-length audio measurably at the training channel, the newest family missed most completely. Three engines, not five: three of the families return numerically identical audio. The finding is; this is the figure re-measured under the shipped configuration. | 93.45 % to 100.00 % missed |
| Genuine half of that same corpusThe control. The failure is the generator, not the channel. | 0.13 % false alarms |
| A different vendor's generator, held outAt the shipped threshold, after retraining on one vendor. 93.01 % at a matched 1 % false-alarm point. | 84.62 % missed |
| Cross-vendor transferAt matched 1 % false-alarm operating points. Teaching the detector one vendor buys almost nothing against another. One held-out vendor, so one data point. | 8.9 % of the within-vendor benefit |
| Read speech, different source familyFull-length clips at the 1 % in-domain operating point. | 0.16 % false alarms |
| Spontaneous speech, consumer microphonesSame threshold, same measurement. The ordering across corpora is unexplained: acoustic distance from the training channel was tested as the cause and rejected. | 20.51 % false alarms |
| ASVspoof 2019 LA evaluation setAll 71,237 trials under the shipped configuration. Computed here under a published definition and positioned against a published column; nobody else scored it. The comparable system on that column, the only other one not trained on that corpus, is at 5.833 %. | 19.391 % EER |
| Genuine one-second clips, zero-paddedThe previous preprocessing. Speech content held fixed, the fill the only variable, so this is causal rather than correlational. | 92.40 % falsely flagged |
| The same clips, tiled to fill the windowWhat ships. Audio under one second is refused rather than scored. | 2.67 % falsely flagged |
| Single-request scoring latencyBatch 1, warm, server-side: read, resample, encoder and head, excluding network. Measured on the deployment hardware over 20 requests, 27.7 to 28.2 ms. | 27.9 ms median |
| Audio it refuses rather than guessing onA detector that returns a number for input it cannot judge is worse than one that declines, because the number will be used. | under 1.0 s, silent, below 8 kHz |