Rich color comes from the temple of Wofei
Qubits | Official account QbitAI
A disc the size of a coin is attached to the throat, and the person with a throat problem can regain a new "voice".
This is the latest research result published by Tsinghua University in the journal Nature, a wearable throat made of graphene material.
△ Nature Machine Intelligence
It can accurately identify subtle vibrations and blurred words in the wearer's throat, and then synthesize them into normal speech, with an average recognition accuracy rate of 99.05%.
It is said that this device is also good in places with very high noise.
In this way, those who cannot speak normally, including laryngectomy patients, temporary throat inflammation, teachers who give lectures for a long time, people who work in a noisy environment but need to communicate, etc., are saved.
Of course, there are many people who directly cue Teacher Li Xuejian.
Tsinghua wearable smart larynx, patient measurement accuracy rate of 91%
To help people with throat problems solve communication problems, scientists have long been studying.
However, some previous solutions are often invasive and not portable due to a range of peripherals or multichannel electrodes.
To solve these two problems, the sensor needs to be sensitive and small enough to be used externally.
It also needs to fit the skin adequately.
Because theoretically, the vibration in the larynx reflects the movement of the vocal cords and related muscle groups.
Some people with vocal vocal disorders may train the esophagus to make sound, so flexible sensors are needed to maintain a fit to the skin to take care of this area.
Fortunately, there is such a thing that can meet these needs.
It is a mechanical sensor made of laser-scribed graphene (LSG), specifically for the surface of the body.
However, because it is uncertain whether the device is sensitive to low-frequency muscle movements and sound vibrations transmitted to the surface of the skin, the authors' team optimized its honeycomb microstructure.
Finally, it can show uniform sensitivity after > 1000 bending tests, accurately identifying vibration information in the frequency range between 100–20kHz.
Since LSG films are electrically and thermally conductive, such devices can also generate sound through thermoacoustic effects. In experiments on sound emission stability, the authors demonstrated that it could remain stable for three hours straight.
Here's how this wearable smart throat works:
First, the device encodes the collected vibration information into a standard pulse code modulation (PCM) signal.
The corresponding spectrogram is then generated by the Fast Fourier transform (FFT).
Then, the model is externally processed and recognized, and the detected multimodal signal is converted into the corresponding speech.
Finally, the thermoacoustic effect drives the device to sound to help the wearer complete the communication.
Experiments have shown that the device's recognition accuracy of phonemes, tones and words reaches an average of 99.05%.
The figure below shows the accuracy of the device at different intensities (dB) of noise.
It can be seen that it can also maintain recognition ability in the case of ambient noise exceeding 60dB, and the noise immunity is far better than that of microphones.
"Talk is cheap", the author also conducted a combat test.
They recruited a volunteer who had completed a laryngectomy (not a total incision) to test the recognition of six daily phrases.
As can be seen from the spectrogram, the wearable smart larynx can sense the vocal vibration of the patient's throat.
However, due to the incomplete vocal organs, people sometimes swallow sounds while speaking. However, the fine-tuning model can still extract enough information from the signal to achieve a recognition accuracy of 81.25%.
Next, they optimized the single AlexNet model used (Alex Net+ReliefF+SVM) and finally achieved 91% recognition accuracy.
Overall, due to the feasible manufacturing process, high sensitivity, stable performance, strong noise immunity, and integrated vocal capabilities, the authors believe that this wearable throat can be an ideal tool for the next generation of speech recognition and interaction systems.
And netizens also opened their brains:
There are filters that allow you to add a timbre adjustment to become a wearable voice changer;
There is also a real-time translation that allows people to directly have the ability to speak multiple languages.
What else do you think is useful?
Paper Address:
https://www.nature.com/articles/s42256-023-00616-6
Reference Links:
https://weibo.com/1231317854/MwsEvkugi?refer_flag=1001030103_
— End —
Qubits QbitAI · Headline number signed
Follow us and be the first to know the latest scientific and technological trends