My Yandex.Station Mini can output two sounds simultaneously, but yours? (UPD. How to command a female voice)

Recently acquired Yandex.Station Mini. If, who does not know, this is a small smart speaker controlled by voice and gestures. Inside is Alice's voice assistant: she turns on the music, answers questions and runs errands. It was acquired as a "smart radio" for the kitchen, the subsequent creation of a smart home with its own skills.







After "pairing" with the operating systems Winodows 7, 10, Ubuntu 16.04, having played a lot with its capabilities, I got acquainted with the official documentation.

, Bluetooth — :



: «, Bluetooth» , .



Bluetooth , Bluetooth.



.



Bluetooth, . , , .

After reading it, I was very upset. On the one hand, the amazing possibilities of speech control, skill creation, smart home. On the other hand, using the charging capabilities from the USB 3.0 port of a laptop, we get a voice control panel with a range of a good Wi-Fi point, without any special tweaks - up to 100 meters!



I decided to check the passphrase on a speaker paired via Bluetooth with a computer. “Alice, turn on the radio Mayak. And the radio began to sing ... And in parallel was the sound from the computer. As a result, the official documentation has been refuted, the device has received new features since August 2020. For two streams, it is recommended to command one stream as usual, using Alice's commands, the stream via Bluetooth can be controlled by the computer volume controls, also using a wireless keyboard with control keys, or, if available, by separate laptop volume control buttons from the end of the device.



Who else can do this experiment?



UPD1. To analyze why there are problems with female voices when controlling a smart speaker, I will give several graphs.



This is a graph of the average power of 20 Russian speakers. As you can see, speech is very uneven, spectral density is concentrated in the region of 200 ... 600 Hertz.

(Fig. taken from "Educational materials OKSO 210000. Electronic engineering, radio engineering and communications. Lectures for teachers and university students." 3. Speech formation and speech characteristics)



Now let's see how the smart speaker listens to us.





(Taken from the video, YouTube user Prokhor Ponomarev , post Measuring the frequency response using the iPad, against the Behringer ECM8000 .

We are interested in the blue curve. This is the frequency response of the iPad 4 microphone, in principle, this is the standard that a smart speaker could strive for. But this , most likely, for such a price they simply did not invest in it programmatically.



From this the author concludes. Look carefully at the graphs: men can speak a command in the lower range of speech, and, due to the greater exhaled mass, “breathe more evenly” when giving the command. In women, the average frequency of the voice is higher, the presence of a greater number of inharmonic sounds does not allow the sensitivity of an array of 4 microphones to reach a uniform plateau. Hence, a simple everyday conclusion - for a woman's voice it is necessary to turn off emotions, try to speak "with the male energy" of the bass, dropping the lower formant.



A number of indirect voice measurements were also carried out, on which the recognition system was trained. So far, the general conclusion is that she trained on male voices.



Under the spoiler, there are several technical screenshots of connecting Yandex.Station Mini to Windows.