HK40112939A - Detection of silent speech - Google Patents

Description

Silent speech detection

Cross-references to related applications

This application claims the benefit of U.S. Provisional Patent Application 63/229,091, filed August 4, 2021, which is incorporated herein by reference.

Invention Field

This invention relates generally to physiological sensing, and more particularly to methods and apparatus for sensing human speech.

background

Speaking activates nerves and muscles in the chest, neck, and face. Therefore, electromyography (EMG) has been used, for example, to capture muscle impulses for speech sensing.

Secondary speckle patterns have been used to monitor movement of skin on the human body. Secondary speckle typically appears in the diffuse reflection of a laser beam from a rough surface, such as skin. By tracking the temporal and amplitude variations of secondary speckle generated by reflections from human skin when irradiated by a laser beam, researchers have measured blood pulse pressure and other vital signs. For example, U.S. Patent 10,398,314 describes a method for monitoring the condition of a subject's body using image data indicating a sequence of speckle patterns generated by the body.

Overview

The embodiments of the present invention described below provide novel methods and devices for sensing human speech.

According to an embodiment of the present invention, a sensing device is also provided, comprising a support and an optical sensing head. The support is configured to fit the ear of a user of the device, and the optical sensing head is held by the support in a position close to the user's face and configured to sense light reflected from the face and output a signal in response to the detected light. A processing circuit is configured to process the signal to generate voice output.

In one embodiment, the support includes an ear clip. Alternatively, the support includes an eyeglass frame. In the disclosed embodiment, an optical sensing head is configured to sense light reflected from the user's cheek.

In some embodiments, the optical sensing head includes an emitter and a sensor array, the emitter being configured to direct coherent light toward a face, and the sensor array being configured to sense a secondary speckle pattern resulting from reflections of the coherent light from the face. In the disclosed embodiments, the emitter is configured to direct multiple beams of coherent light to different corresponding locations on the face, and the sensor array is configured to sense secondary speckle patterns reflected from these locations. Additionally or alternatively, the locations illuminated by the beams and sensed by the sensor array extend over an area of at least 1 ^cm² . Furthermore, additionally or alternatively, the optical sensing head includes a plurality of emitters configured to generate corresponding beam sets covering different, corresponding areas of the face, and processing circuitry is configured to select and actuate a subset of the emitters, without actuating all emitters.

In the disclosed embodiments, the processing circuitry is configured to detect changes in the sensed secondary speckle pattern and generate speech output in response to the detected changes.

Alternatively or additionally, the processing circuitry is configured to operate the sensor array at a first frame rate, sense facial motion in response to a signal when operating at the first frame rate, and increase the frame rate to a second frame rate greater than the first frame rate in response to the sensed motion, in order to generate speech output.

In the disclosed embodiments, the processing circuitry is configured to generate voice output in response to changes in the signal output by the optical sensing head caused by movement of the user's skin surface without the user making any sound.

Typically, the optical sensing head is held by a bracket at a distance of at least 5 mm from the user's skin surface.

In one embodiment, the device includes one or more electrodes configured to contact the user's skin surface, wherein processing circuitry is configured to generate voice output in response to electrical activity sensed by the one or more electrodes and signals output from an optical sensing head.

Additionally or alternatively, the device includes a microphone configured to sense sounds emitted by a user. In one embodiment, processing circuitry is configured to compare a signal output from the optical sensing head with the sound sensed by the microphone to calibrate the optical sensing head. Additionally or alternatively, the processing circuitry is configured to change the operating state of the device in response to sensing sounds emitted by the user.

In some embodiments, the device includes a communication interface, wherein processing circuitry is configured to encode signals for transmission via the communication interface to a processing device, which processes the encoded signals to generate voice output. In the disclosed embodiments, the communication interface includes a wireless interface.

Additionally or alternatively, the device includes a user control connected to a bracket and configured to sense gestures made by a user, wherein processing circuitry is configured to change the operating state of the device in response to the sensed gestures.

Additionally or alternatively, the device includes a speaker configured to fit inside a user's ear, wherein processing circuitry is configured to synthesize an audio signal corresponding to the speech output for playback by the speaker.

According to an embodiment of the present invention, a sensing method is also provided, the method comprising sensing movement of skin on the face of a human object in response to the human object articulating a word without audibly pronouncing the word. In response to the sensed movement, a speech output including the spoken word is generated.

In some embodiments, sensing the motion includes sensing light reflected from the face of the object. In the disclosed embodiments, sensing the light includes directing coherent light to the skin and sensing a secondary speckle pattern resulting from the reflection of the coherent light from the skin. In one embodiment, directing the coherent light includes directing multiple beams of coherent light to different corresponding locations on the face and using a sensor array to sense the secondary speckle pattern reflected from each location.

In the disclosed embodiments, generating speech output includes synthesizing an audio signal corresponding to the speech output. Alternatively or additionally, generating speech output includes transcribing words spoken by the subject.

The invention will be more fully understood from the following detailed description of embodiments thereof, taken in conjunction with the accompanying drawings, in which:

Brief description of the attached diagram

Figure 1 is a schematic illustration of a system for voice sensing according to an embodiment of the present invention;

Figure 2 is a schematic cross-sectional view of an optical sensing head according to an embodiment of the present invention;

Figure 3 is a schematic illustration of a voice sensing device according to another embodiment of the present invention;

Figure 4 is a block diagram schematically illustrating the functional components of a system for voice sensing according to an embodiment of the present invention; and

Figure 5 is a flowchart schematically illustrating a voice sensing method according to an embodiment of the present invention.

Detailed Implementation

People communicate almost anytime, anywhere via their mobile phones. The widespread use of mobile phones in public places introduces discordant noise and often raises privacy concerns, as conversations are easily overheard by passersby. Furthermore, when one party in a phone conversation is in a noisy environment, the other party or parties may have difficulty understanding what they are hearing due to background noise. Text communication offers a solution to these problems, but text input on mobile phones is slow and interferes with users' ability to see where they are going.

The embodiments of the invention described herein address these problems using silent speech, enabling users to speak words and sentences without actually uttering them or making any sound at all. Normal phonation uses multiple groups of muscles and nerves, starting in the chest and abdomen, passing through the throat, and rising through the mouth and face. To pronounce a given phoneme, motor neurons activate muscle groups in the face, throat, and mouth to prepare for airflow from the lungs, and these muscles continue to move during speech to create words and sentences. Without this airflow, the mouth does not produce sound. Silent speech occurs when there is no airflow from the lungs, but the muscles in the face, throat, and mouth continue to produce the desired sound.

Silent speech can be caused by neurological and muscular disorders; however, it can also occur intentionally, such as when we say words without intending to be heard. This speaking can even occur when we conceptualize spoken words without opening our mouths. The resulting activation of our facial muscles causes subtle movements on the skin surface. The inventors have discovered that by properly sensing and decoding these movements, it is possible to reliably reconstruct the actual sequence of words spoken by the user.

Therefore, embodiments of the invention described herein sense subtle movements of the skin and subcutaneous nerves and muscles on a subject's face and use the sensed movements to generate a speech output including spoken words, the subtle movements occurring in response to the words spoken by the subject, whether phonologically or nonverbally. These embodiments provide methods and apparatus for sensing these subtle movements without contact with the skin (e.g., by sensing light reflected from the subject's face). Thus, they enable users to communicate silently with others or record their own thoughts in a manner substantially imperceptible to the other party. The devices and methods according to these embodiments are also insensitive to ambient noise and can be used in virtually any environment without requiring the user to divert their gaze and attention from their surroundings.

Some embodiments of the present invention provide sensing devices in the form of common consumer products, such as clip-on headphones or glasses. In these embodiments, an optical sensing head is held close to the user's face by a bracket adapted to fit inside or above the user's ear. For example, by directing coherent light to an area of the face (e.g., the cheek), the optical sensing head senses light reflected from the face and senses changes in the secondary speckle pattern resulting from the reflection of coherent light from the face. Processing circuitry in the device processes the signal output by the optical sensing head due to the reflected light to generate corresponding voice output.

Alternatively, the principles of the invention can be implemented without ear clips or other supports. For example, in an alternative embodiment, a silent voice sensing module including a coherent light source and sensors can be integrated into a mobile communication device such as a smartphone. This integrated sensing module senses silent voice when the user holds the mobile communication device in a suitable position close to their face.

As used in this specification and claims, the term "light" refers to electromagnetic radiation in any or all ranges of the infrared, visible, and ultraviolet ranges.

Figure 1 is a schematic illustration of a system 18 for voice sensing according to an embodiment of the present invention. System 18 is based on a sensing device 20, wherein a bracket in the form of an ear clip 22 is adapted to the ear of a user 24 of the device. An earphone 26 attached to the ear clip 22 is adapted to fit inside the user's ear. An optical sensing head 28 is connected to the ear clip 22 via an arm 30, thus holding it in a position close to the user's face. In the illustrated embodiment, device 20 has the form and appearance of a clip-on headset, wherein the optical sensing head replaces a microphone (or has an optical sensing head in addition to a microphone).

The optical sensing head 28 directs one or more beams of coherent light to different corresponding locations on the user 24's face, thereby generating an array of light spots 32 extending over a region 34 of the face (specifically, on the user's cheeks). In this embodiment, the optical sensing head 28 does not contact the user's skin at all, but is held at a certain distance from the skin surface. Typically, this distance is at least 5 mm, and it can even be larger, for example, at least 1 cm or even 2 cm or more from the skin surface. In order to be able to sense the movement of different parts of the facial muscles, the region 34 covered by the light spots 32 and sensed by the optical sensing head 28 typically has a range of at least 1 ^cm² ; and a larger region (e.g., at least 2 ^cm² or even greater than 4 ^cm² ) can be advantageous.

Optical sensing head 28 senses coherent light reflected from spot 32 on the face and outputs a signal in response to the detected light. Specifically, optical sensing head 28 senses the secondary speckle pattern generated by the reflection of coherent light from each spot 32 within its field of view. To cover a sufficiently large area 34, this field of view typically has a wide angular range, typically at least 60°, or possibly 70°, or even 90° or more. Within this field of view, device 20 can sense and process the signal generated by the secondary speckle pattern of all spots 32 or only a subset of spots 32. For example, device 20 can select a subset of spots that has been found to provide the greatest amount of useful and reliable information regarding the relevant movements of the user 24's skin surface. The details of the structure and operation of optical sensing head 28 are described below with reference to FIG2.

Within system 18, processing circuitry processes the signal output from optical sensing head 28 to generate speech output. As previously described, even if user 22 does not utter a sound or make any other noise, the processing circuitry is able to sense the movement of user 22's skin and generate speech output. The speech output may take the form of a synthesized audio signal, a text transcription, or both. The synthesized audio signal can be played back via a speaker in headphones 26 (and is useful in providing feedback to user 22 regarding the speech output). Additionally or alternatively, the synthesized audio signal may be transmitted over a network, for example via a communication link with a mobile communication device (e.g., smartphone 36).

The processing circuitry in system 18 can function entirely within device 20, or alternatively, it can be distributed between device 20 and an external processor, such as the processor in smartphone 36 running suitable application software. For example, the processing circuitry within device 20 can digitize and encode the signal output from optical sensing head 28 and transmit the encoded signal to smartphone 36 via a communication link. This communication link can be wired or wireless, for example, using the Bluetooth ^™ wireless interface provided by the smartphone. The processor in smartphone 36 processes the encoded signal to generate voice output. Smartphone 36 can also access server 38 via a data network such as the Internet to, for example, upload data and download software updates. Details of the design and operation of the processing circuitry are described below with reference to FIG4.

In the illustrated embodiment, device 20 also includes a user control 35, in the form of a push-button sensor or proximity sensor, which is connected to ear clip 22. User control 35 senses gestures performed by the user, such as pressing on user control 35 or otherwise bringing the user's finger or hand close to user control. In response to an appropriate user gesture, processing circuitry changes the operating state of device 20. For example, user 24 can switch device 20 from idle mode to active mode in this way, and thus signal that the device should begin sensing and generating voice output. This switching is useful in terms of saving battery power in device 20. Alternatively or additionally, other methods can be applied to control the operating state of device 20 and reduce unnecessary power consumption, such as those described below with reference to FIG5.

Figure 2 is a schematic cross-sectional view of the optical sensing head 28 of the device 20, illustrating the components and functional details of the optical sensing head according to an embodiment of the present invention. The optical sensing head 28 includes a transmitter module 40 and a receiver module 48, as well as an optional microphone 54.

The transmitter module 40 includes a light source, such as an infrared laser diode 42, that emits an input beam of coherent radiation. A beam-splitting element 44, such as a Dammann grating or another suitable type of diffractive optics (DOE), splits the input beam into a plurality of output beams 46, which form corresponding spots 32 at a position matrix extending over region 34. In one embodiment (not shown in the figures), the transmitter module 40 includes a plurality of laser diodes or other transmitters that produce corresponding sets of output beams 46 that cover different corresponding sub-regions within region 34 of the user's face. In this case, the processing circuitry in device 20 can select and actuate only a subset of the transmitters, without actuating all of them. For example, to reduce the power consumption of device 20, the processing circuitry can actuate only one transmitter or a subset of two or more transmitters that illuminates a region on the user's face that has been found to provide the most useful information for generating the desired speech output.

Receiver module 48 includes an array 52 of optical sensors, such as a CMOS image sensor, wherein objective lens 50 is used to image region 34 onto array 52. Due to the small size of optical sensing head 28 and its proximity to the skin surface, as described above, receiver module 48 has a sufficiently wide field of view and observes numerous light spots 32 at high angles away from the normal. Due to the roughness of the skin surface, secondary speckle patterns at these light spots 32 can also be detected at these high angles.

Microphone 54 senses the voice emitted by user 24, allowing user 22 to use device 20 as a conventional headset when needed. Additionally or alternatively, microphone 54 can be used in conjunction with the silent voice sensing capability of device 20. For example, microphone 54 can be used during calibration, where optical sensing head 28 senses skin movement as user 22 utters certain phonemes or words. Processing circuitry can then compare the signal output from optical sensing head 28 with the sound sensed by microphone 54 to calibrate the optical sensing head. This calibration may include prompting user 22 to move the position of optical sensing head 28 to align the optical components at a desired location relative to the user's cheek.

In another embodiment, the audio signal output by microphone 54 can be used to change the operating state of device 20. For example, the processing circuit can generate speech output only when microphone 54 does not detect the user 24 uttering a word. Other applications of the combination of optical sensing and acoustic sensing provided by optical sensing head 28 and microphone 54 will be apparent to those skilled in the art upon reading this specification and are considered to be within the scope of the invention.

Figure 3 is a schematic illustration of a voice sensing device 60 according to another embodiment of the present invention. In this embodiment, the ear clip 22 is integrated with or otherwise attached to an eyeglass frame 62. A nasal electrode 64 and a temporal electrode 66 are attached to the frame 62 and contact the user's skin surface. Electrodes 64 and 66 receive surface electromyography (sEMG) signals, which provide additional information about the activation of the user's facial muscles. The processing circuitry in the device 60 uses the electrical activity sensed by electrodes 64 and 66 and the output signal from the optical sensing head 28 to generate speech output from the device 60.

Additionally or alternatively, device 60 includes one or more additional optical sensing heads 68, similar to optical sensing head 28, for sensing skin movement in other areas of the user's face. These additional optical sensing heads may be used in conjunction with or in place of optical sensing head 28.

Figure 4 is a block diagram schematically illustrating the functional components of a system 18 for voice sensing according to an embodiment of the present invention. The illustrated system is constructed around the components shown in Figure 1, including a sensing device 20, a smartphone 36, and a server 38. Alternatively, the functions shown in Figure 4 and described below may be implemented and distributed differently among the components of the system. For example, some or all of the processing capabilities belonging to the smartphone 36 may be implemented in the sensing device; or the sensing capabilities of the device 20 may be implemented in the smartphone 36.

In the illustrated example, as described above, sensing device 20 includes a transmitter module 40, a receiver module 48, a speaker 26, a microphone 54, and a user control (UI) 35. For completeness, sensing device 20 is shown in FIG. 4 as also including other sensors 71, such as electrodes and/or environmental sensors; however, as previously stated, sensing device 20 is capable of operating solely based on non-contact measurements performed by the transmitter and receiver modules.

The sensing device 20 includes processing circuitry in the form of an encoder 70 and a controller 75. The encoder 70 includes hardware processing logic and/or a digital signal processor (DSP), which may be hardwired or programmable. The DSP extracts and encodes features from the output signal from the receiver module 48. The sensing device 20 transmits the encoded signal to a corresponding communication interface 77 in the smartphone 36 via a communication interface 72, such as a Bluetooth interface. A battery 74 provides operating power to the components of the sensing device 20.

The controller 75 includes a programmable microcontroller, which, for example, sets the operating state and operating parameters of the sensing device 20 based on inputs received from the user control 35, the receiver module 48, and the smartphone 36 (via the communication interface 72). Some aspects of this functionality are described below with reference to FIG5. In an alternative embodiment, the controller 75 includes a more powerful microprocessor and/or processing array, independent of the smartphone 36, which locally processes the characteristics of the output signal from the receiver module 48 and generates voice output within the sensing device.

However, in this embodiment, the encoded output signal from the sensing device 20 is received in the memory 78 of the smartphone 36 and processed by a speech generation application 80 running on the processor in the smartphone 36. The speech generation application 80 converts features in the output signal into a sequence of words in the form of text and/or audio output signals. The communication interface 77 transmits the audio output signal back to the speaker 26 of the sensing device 20 for playback to the user. The text and/or audio output from the speech generation application 80 is also input to other applications 84, such as voice and/or text communication applications and recording applications. The communication applications communicate, for example, via a cellular or Wi-Fi network through a data communication interface 86.

The operation of encoder 70 and speech generation application 80 is controlled by local training interface 82. For example, interface 82 can instruct encoder 70 which temporal and spectral features to extract from the signal output by receiver module 48, and can provide speech generation application 80 with coefficients of a neural network that converts these features into words. In this example, speech generation application 80 implements an inference network that finds the word sequence with the highest probability corresponding to the encoded signal features received from sensing device 20. Local training interface 82 receives coefficients from inference network from server 38, which can also periodically update the coefficients.

To generate local training instructions 82, server 38 uses data repository 88, which contains speckle images and corresponding ground truth spoken words from the set of training data 90. Repository 88 also receives training data collected in the field from sensing device 20. For example, training data may include signals collected from sensing device 20 when a user speaks certain sounds and words (potentially including silent and spoken speech). This combination of general training data 90 and individual training data received from each user at each sensing device 20 enables server 38 to derive optimal inference network coefficients for each user.

Server 38 uses image analysis tool 94 to extract features from speckle images in repository 88. These image features, along with a corresponding word dictionary 104 and language model 100, are input as training data into neural network 96. Language model 100 defines the phonetic structure and syntactic rules of the specific language used in the training data. Neural network 96 generates optimal coefficients for inference network 102, which converts the input sequence of the feature set extracted from the corresponding sequence of speckle measurements into corresponding phonemes and ultimately into a sequence of words. Further details of the network architecture and training process are described in the aforementioned provisional patent application. Server 38 downloads the coefficients of inference network 102 to smartphone 36 for use in speech generation application 80.

Figure 5 is a flowchart schematically illustrating a method for voice sensing according to an embodiment of the present invention. For convenience and clarity, the method is described with reference to the elements of system 18 shown in Figures 1 and 4 and described above. Alternatively, the principle of the method can be applied in other system configurations, such as a system configuration using sensing device 60 (Figure 3) or a sensing device integrated into a mobile communication device.

In idle step 110, as long as user 24 is not speaking, sensing device 20 operates in low-power idle mode to conserve power in battery 74. In this mode, controller 75 drives the array 52 of sensors in receiver module 48 at a low frame rate (e.g., 20 frames/second). Transmitter module 40 may also operate at reduced output power. In motion detection step 112, while receiver module 48 operates at this low frame rate, controller 75 processes the image output from array 52 to detect facial movements indicative of speech. In motion capture step 114, when such motion is detected, controller 75 instructs receiver module 48 and other components of sensing device 20 to increase the frame rate to, for example, a range of 100-200 frames/second to detect changes in the secondary speckle pattern due to silent speech. Alternatively or additionally, controller 75 may increase the frame rate and power other components of sensing device 20 in response to other inputs, such as actuation of user control 35 or instructions received from smartphone 36.

The image captured by receiver module 48 typically contains a matrix of the projected laser spots 32, as shown in Figure 1. In spot detection 116, encoder 70 detects the location of the spots in the image. The encoder can extract features from all spots; however, to save power and processing resources, it is desirable for the encoder to select a subset of spots. For example, local training interface 82 can indicate which subset of spots contains the maximum amount of information about the user's speech, and encoder 70 can select spots from that subset. In cropping step 118, encoder 70 crops small windows from each image, where each such window contains one of the selected spots.

In feature extraction step 120, encoder 70 extracts features of speckle motion from each selected speckle. For example, encoder 70 can estimate the total energy in each speckle based on the average intensity of pixels in the corresponding window, and can measure the change in energy of each speckle over time. Additionally or alternatively, encoder 70 can extract other temporal and/or spectral features of speckles in the selected subset of speckles. Encoder 70 transmits these features to speech generation application 80 (running on smartphone 36), where, in feature input step 122, speech generation application 80 inputs a vector of feature values into inference network 102 downloaded from server 38.

In the speech output step 124, based on the sequence of feature vectors input to the inference network over time, the speech generation application 80 outputs a stream of words, which are concatenated into sentences. As previously described, the speech output is used to synthesize audio signals for playback via speaker 26. In the post-processing step 126, another application 84 running on smartphone 36 post-processes the speech and/or audio signals to record corresponding text and/or transmit speech or text data over the network.

It should be understood that the above embodiments are cited by way of example, and the present invention is not limited to what has been specifically shown and described above. More precisely, the scope of the present invention includes combinations and sub-combinations of the various features described above, as well as variations and modifications of these features that would occur to those skilled in the art after reading the foregoing description and that are not disclosed in the prior art.

Claims (30)

1. A sensing device, comprising: A stand, the stand being configured to fit the ear of a user of the device; An optical sensing head, held by the bracket in a position close to the user's face, is configured to sense light reflected from the face and output a signal in response to the detected light; and A processing circuit configured to process the signal to generate voice output. 2. The device according to claim 1, wherein the bracket includes an ear clip. 3. The device according to claim 1, wherein the bracket comprises an eyeglass frame. 4. The device of claim 1, wherein the optical sensing head is configured to sense light reflected from the user's cheek. 5. The device of claim 1, wherein the optical sensing head includes an emitter and a sensor array, the emitter being configured to direct coherent light to the face, and the sensor array being configured to sense a secondary speckle pattern resulting from reflection of the coherent light from the face. 6. The device of claim 5, wherein the transmitter is configured to direct a plurality of beams of the coherent light to different corresponding locations on the face, and the sensor array is configured to sense secondary speckle patterns reflected from said locations. 7. The device of claim 6, wherein the position illuminated by the light beam and sensed by the sensor array extends over a field of view having an angular width of at least 60°. 8. The device of claim 6, wherein the position illuminated by the light beam and sensed by the sensor array extends over an area of at least 1 ^cm² . 9. The device of claim 6, wherein the optical sensing head includes a plurality of emitters configured to generate corresponding groups of the light beams, the corresponding groups of the light beams covering different corresponding regions of the face, and wherein the processing circuitry is configured to select and actuate a subset of the emitters without actuating all of the emitters. 10. The device of claim 5, wherein the processing circuitry is configured to detect a change in the sensed secondary speckle pattern and generate the speech output in response to the detected change. 11. The device of claim 5, wherein the processing circuitry is configured to operate the sensor array at a first frame rate, sense movement of the face in response to the signal when operating at the first frame rate, and increase the frame rate to a second frame rate greater than the first frame rate in response to the sensed movement to generate the speech output. 12. The device according to claims 1-11, wherein the processing circuitry is configured to generate the voice output in response to a change in the signal output by the optical sensing head caused by movement of the user's skin surface without the user making any sound. 13. The device according to claims 1-11, wherein the optical sensing head is held by the bracket at a position at least 5 mm away from the user's skin surface. 14. The device of claims 1-11, further comprising one or more electrodes configured to contact a user's skin surface, wherein the processing circuitry is configured to generate the voice output in response to electrical activity sensed by the one or more electrodes and the signal output by the optical sensing head. 15. The device according to claims 1-11, further comprising a microphone configured to sense sounds emitted by a user. 16. The device of claim 16, wherein the processing circuitry is configured to compare the signal output by the optical sensing head with the sound sensed by the microphone in order to calibrate the optical sensing head. 17. The device of claim 16, wherein the processing circuitry is configured to change the operating state of the device in response to sensing a sound emitted by a user. 18. The device according to claims 1-11, further comprising a communication interface, wherein the processing circuitry is configured to encode the signal for transmission via the communication interface to a processing device, the processing device processing the encoded signal to generate the voice output. 19. The device of claim 17, wherein the communication interface includes a wireless interface. 20. The device of claims 1-11, further comprising a user control connected to the bracket and configured to sense a gesture made by a user, wherein the processing circuitry is configured to change the operating state of the device in response to the sensed gesture. 21. The device according to claims 1-11, further comprising a speaker configured to fit within a user's ear, wherein the processing circuitry is configured to synthesize an audio signal corresponding to the speech output for playback by the speaker. 22. A sensing method, comprising: In response to a human object saying a word but not uttering the word aloud, and sensing the movement of the skin on the object's face without contact; and In response to sensed motion, a speech output is generated, which includes the spoken words. 23. The method of claim 23, wherein sensing the motion includes sensing light reflected from the face of the object. 24. The method of claim 24, wherein sensing the light comprises directing coherent light to the skin and sensing a secondary speckle pattern resulting from the reflection of the coherent light from the skin. 25. The method of claim 25, wherein guiding the coherent light comprises guiding a plurality of beams of the coherent light to different corresponding locations on the face and using a sensor array to sense secondary speckle patterns reflected from each location. 26. The method of claim 26, wherein the position illuminated by the light beam and sensed by the sensor array extends over a field of view having an angular width of at least 60°. 27. The method of claim 26, wherein the location, illuminated by the light beam and sensed by the sensor array, extends over an area of at least 1 ^cm² on the cheek of the subject. 28. The method of claim 25, wherein generating the speech output includes detecting a change in the sensed secondary speckle pattern and generating the speech output in response to the detected change. 29. The method according to any one of claims 23-29, wherein generating the speech output comprises synthesizing an audio signal corresponding to the speech output. 30. The method according to any one of claims 23-29, wherein generating the speech output comprises transcribing words spoken by the subject.