Dlexa: Hedge Dragon Using Mycroft AI

Getting Dlexa to Speak to You

This page is about giving Dlexa a voice. It could be about buying an expensive weather tight speaker and mounting on a stick but this is an experiment so even her voice needs to be DIY using a rare technology. The choice was to build a DML Distributed Mode Loudspeaker.

Quick Test:

To test the DML using a music player (VLC) click the link in this web page sends an audio file (mp3) out a USB sound fob amplified on a Class-D amp so it plays on the DML speaker.
Remote VLC to Dragon: [is currently disabled]

Testing

May 2023 tests showed the chain of electronics from the linux server through a USB sound card to a Class-D amplifier driving exciters on the foam (Formular) DMP speaker in the DML frame works.

Trouble following the previous run-on sentence? It is just to show "working" means the chain of roughly a dozen hardware and software parts must all do their jobs to get music to play on the speaker. Working means music playing on the DML speaker can be controlled from any remote PC or cell phone.

Using a single 3 watt DML exciter (center mount) on the 24"x24" DML plate proved too quiet at 100% volume and distorted at 120%. A second exciter installed off center (2/5 location) on the plate provided an excellent sound even at 80% volume.

Exciter placement is the subject of much debate with many insisting a exciter should never be placed at the center because it will cause a frequency peak. Off center placement means more complex resonance pattern and a flat frequency response. In practice the first exciter at center (12" x 12") and the second offset at the 2/5 by 3/5 point (9.6" x 14.4") works just fine. One note: make sure you have the polarity right. If the exciters are wired properly their power will sum. If you get it wrong the sound with be the difference: not zero because of the placement but significantly quieter than one exciter by itself.

While the ears (microphone) worked (3 days) Dlexa was able to answer questions so the ears+mind+voice worked from end to end. The DML frame is periodically tested and has been completely reliable over the summer. Work now is focusing on the more delicate and troublesome ears.

Area Versus Point Source

A standard speaker is a "point source" of sound and like a single light bulb it is easy to identify its location. Filling a space with sound means using many speakers in specially constructed boxes so a listener at a distance hears and average sound. The "perfect" average is usually a small circle quite distant from the speakers array.

A DML is an "area source" of sound like a sheet of concrete in the sun. There is a point source (the sun) but a photo of the concrete just sees a sheet of uniform light. For a DML the exciter is a point source (like the sun) but it moves the (Formular) plate and the sound comes from the entire four square feet of the plate (like the concrete sheet). The result is the "perfect" listening spot is starts about 4 feet in front of the plate and is a roughly rectangular region about 12 feet by 12 feet.

The DML effect is quite magical. When you walk through the 12 foot x 12 foot space the sound appears to come from everywhere. It gets a little louder near the center of that space but not enough for you to identify the source. You know it has something to do with the grey-pink plate behind the bug screen but it does not appear to be the "source".

The disadvantage of a DML speaker is you cannot get it to throb like a sub-woofer so you would still need one of those for a dance floor. It is however perfect for Dlexa's voice because the sound fills the space where the subject stands before the dragon.

Next

Once the "sound infrastructure" is in place the linux server running Mycroft can speak. That done: what Dlexa and Pearl say becomes (just?) a programming problem.

The next task is to install the permanent wiring (under the hedge body) for power, WiFi, and USB to the DML Frame. See Hedge for the cabling plans.