VoxTexto.Listen, compare and review a real transcript
Listen to the actual test file, inspect each mode's output and download an editorially reviewed version. These are not testimonials or private recordings.
1. The public recording
This short Spanish reading comes from the open FLEURS dataset. Download the WAV to repeat the test. You do not need to select any personal file.
2. The same voice with added noise
We added artificial white noise to a copy to see how recognition changes. This is not a cleanup filter or a recording made in a real noisy location. The source voice and spoken content are unchanged.
3. Unedited automatic output
| Audio | Mode | Measured time | Unedited automatic result |
|---|---|---|---|
| Clear | Fast / Base | 5.8 s | Milenios atrás, un hombre llamado Aristarcho comentó que el sistema solar gira a mentor al sol. |
| Added noise | Fast / Base | 5.9 s | Y el año atrás un hombre llamado Aristarco comentó que hiciste más su hogar y dado en torno al sol. |
| Clear | Higher accuracy / Small | 15.2 s | Milenios atrás, un hombre llamado Aristarco, comentó que el sistema solar giraba en torno al sol. |
| Added noise | Higher accuracy / Small | 15.0 s | Milenio, atrás, un hombre llamado Alistar Co. comentó que el sistema solar giraba en torno al sol. |
4. Editorially reviewed reference
Milenios atrás, un hombre llamado Aristarco comentó que el sistema solar giraba en torno al Sol.The review checks the spelling of Aristarco and the Spanish expression “en torno al Sol”. A plausible word is not confirmed until checked against the recording. This reference explains editing; it is not a broad model evaluation.
5. Open the sample documents
All three contain the same reviewed Spanish reference. TXT is reusable plain text; Word supports further editing; PDF preserves page layout.
Repeat the test
- Download one of the two samples.
- Open VoxTexto and select that WAV.
- Choose Spanish, CPU and the mode to compare; leave speech detection enabled.
- Listen to the passages, edit the text and keep your result.
How we ran the comparison
We used the same 8.4-second Spanish reading and a derived copy with deterministic artificial white noise. Both modes ran on CPU, with Spanish selected manually and speech detection enabled. We first warmed each model's browser cache. Each measured run starts a new recognition worker and includes initialization and inference, but avoids the first large download. Times are observations on one computer, not promised processing speeds.
Test date: 2026-11-06. Chrome / Windows; CPU; ONNX q8; 16 kHz mono. Read the full measurements.