Voice Bench
How this works
Load a voice from the repo or start a new one, generate takes for each filler and stage the one that you like, then download the zip and attach it to the task.
1Voice which one, and how it is synthesized
2Fillers one row per phrase — generate, compare, stage
| Phrase ships in config.json | Say as spoken instead | Takes pick the one that you like |
|---|
3Notes
Kept in voice.md above the generated filler tables, and round-tripped
verbatim. Task, tuning notes, anything the next person needs.
4Review what ships, and what is wrong with it
How this works
Hear a voice say a line with whatever configs you pick — one from the repo, or an id the repo has never seen. Nothing is saved.
How this works
Say each text several times on one gemini voice, to see which of them it refuses.