Tutorial 25 · Voice in depth: how it speaks and how it hears you

What the video says

1. JMAX's voice is set up here, in the menu, under Management and administration. Everything you hear and everything you say is processed on your own computer: your voice is never sent anywhere.
2. At the top you choose how it sounds. There are eighteen voices, with different accents and timbres, female and male. It is the first thing anyone notices about JMAX, so it is worth trying a few.
3. The speed is set with the slider, from slow to fast. Press Listen to hear that voice at that speed before deciding: you don't have to save in order to try it.
4. Below, the engine says where the voice comes from. The Microsoft voices sound more natural and offer more accents, but they need the internet and take about a second longer per sentence.
5. The local voices, the Kokoro ones, run inside the application: they answer sooner and work offline. If you work without internet, or you want the snappiest reply, pick one of these.
6. Whichever you choose, press Save to keep it. The change applies straight away, on the next sentence it speaks.
7. Now, how it understands you. The first option, Rapido, uses Vosk and writes instantly while you speak. It is the quickest, but it can get proper nouns and unusual words wrong.
8. The second, Preciso, adds Whisper. You still see the text instantly, but when each sentence closes Whisper listens to it again and writes it far better. It takes a little longer and uses more machine.
9. Choosing Preciso reveals the correction quality. The larger models write better but want a graphics card; without one, stay on the recommended setting. Next to it you can see which model is loaded and where it runs.
10. Microphone: how it hears you. If you use a Bluetooth headset, Windows shows it twice; take the one that says hands-free. Search reads the devices again if you have just plugged one in.
11. Speaker: how it speaks to you. Press Test sound to confirm it comes out where you expect, especially after changing headphones.
12. Raise the sensitivity if JMAX cannot hear you. And the Your voice bar is the honest check: speak and see whether it moves. If it doesn't move, the problem is the microphone, not JMAX.
13. If something fails, Test my voice goes through the whole chain and tells you exactly where it breaks, instead of leaving you guessing. It takes a few seconds: when it asks, say out loud hello JMAX, can you hear me.
14. There are seven steps: that the system lets it use the microphone, that the microphone is awake, that your voice comes in strong enough, that the sound reaches JMAX, that it understands you, that it produces the spoken answer, and that the answer plays on your speaker.
15. At the bottom, the status says what is loaded right now: live listening, transcription, the local voice and the Microsoft voices. If something shows as unavailable, that is your clue about what to check.