Voice speed test
Tries three small voice makers on this device and times them. Nothing here changes Benny's Hub. Run it in Chrome, the browser Ben uses.
This device
How to use this page
- Plugged in, on Wi-Fi: press Download & load on each voice card. Note how long each takes.
- Press Run the 3 test sentences on each card. Listen for gaps or stutters.
- Turn Wi-Fi off, reload this page, and run them again (they should still work).
- Unplug, then use Repeat for 10 minutes at the bottom.
- Press Copy results and paste them into a message.
The three sentences:
Pocket TTS: speed and sound settings
Pick a setting, then press Apply. The page reloads, and Pocket needs Download & load again. The setting in use is listed under "This device" and in the copied results.
Pocket TTS: custom voice
Pocket TTS can copy the sound of a short voice clip (up to 10 seconds). Four invented sample voices came with this page, or pick a clip of your own, or a voice file exported from Voice Studio (.pocketvoice.json). Only use a real person's voice with their permission: your own voice, or recordings someone kept of their own voice to speak with after losing it, are exactly what this is for.
Your own sentence
Repeat for 10 minutes
Speaks one sentence again and again to see battery use, heat, and whether it slows down. Unplug first. Note the battery % shown below when it starts and ends.
Results
"First sound" is the wait between pressing the button and hearing the voice start (it starts once 0.3 seconds of speech is ready). Targets: short about 1 second or less, medium 1.5, long 2. No gaps.
| Voice | Sentence | First sound (s) | Finished making (s) | Speech length (s) | Speed (× real time) | Gaps | Result |
|---|
Credits
Pocket TTS by Kyutai (CC BY 4.0), run in the browser with code and ONNX files from KevinAHM's web demo (code Apache 2.0). Voices are only copied with permission, as Kyutai's terms require. Piper voices (MIT) via piper-tts-web. Kokoro (Apache 2.0) via kokoro-js. Runtime: ONNX Runtime Web (MIT).