Free guide · Oct 8, 2026
Run a fair AI narrator listening test
How to make matched samples, mix the same background noise and score the words fairly.
The short version
Use real Pocket TTS and Eleven v4 outputs, the same original text and the same fixed music excerpt. Hide the tool labels and target transcript until the listener answers. Score what they understood before asking which voice they preferred. Keep the clean originals, mixed versions and settings so the result can be reproduced.
This is an experiment protocol, not a completed comparison. It measures the two actual configurations you choose; it cannot establish a universal winner. The Pocket TTS repository and Eleven v4 announcement establish the available speech options. Code, model, voice and music rights must each permit your use.
What's in the full guide
- 01Original test text and answer key
- 02The actual Pocket fixture generated for this guide
- 03Select voice configurations honestly
- 04Keep generation and mixing separate
- 05The assembled Reel comparison
- 06Blind presentation order
- 07Answer sheet and result table
- 08Turn the result into a production decision