Patent Application
20050281412
The invention relates to prosthetic devices, and in particular, to prosthetic devices for replacing the human voice.
The electrolarynx is an electrically powered device that produces a sound (or buzz) that can be used to acoustically excite the vocal tract as a substitute for laryngeal voice production. The most common electrolarynx is a neck-placed electrolarynx. These electrolarynxes transmit sound energy through neck tissue to provide acoustic excitation of the vocal tract.
Most present electrolarynxes use a simple, battery powered electromechanical driver that operates like a piston hitting a drumhead to produce a “buzz-like” sound. Typically, the patient holds the drumhead against the neck and activates an electromechanical driver. This driver forces a small cylindrical head mounted on a diaphragm to repeatedly strike a rigid plastic disk, thus, producing a series of impulse-like excitations.
In a conventional electrolarynx, the patient must use one hand to hold the device against the neck. The need to use one hand to control the electrolarynx is physically limiting since it precludes normal bimanual function, i.e., performing manual tasks that require the use of both hands while talking.
A conventional electrolarynx also produces speech that sounds non-human (mechanical, robotic, monotone), has reduced intelligibility and loudness, and draws undesirable attention to the user. The poor quality of EL speech has been traced to limitations in performance of current EL sound generating transducers, and to the loss of the fine control of pitch, amplitude, and voice onset and offset timing that is normally provided by the laryngeal mechanism. Loss of fine control causes deficits in voice-related segmental (e.g., voiced-voiceless distinctions for consonants) and suprasegmental (e.g., intonation, syllabic stress) speech parameters.
In one aspect, the invention includes a voice prosthesis having a voice actuator for generating a signal to be modulated into speech and a neural interface for receiving a signal indicative of neural activity. A signal processing system in communication with both the neural interface and the voice actuator is configured to provide the voice actuator with a control signal representative of the neural activity.
In some embodiments, the neural interface includes a non-invasive interface. Other embodiments include these in which the neural interface is configured to detect neural activity associated with contraction of neck strap muscles.
Additional embodiments include those in which the voice actuator includes an electrolarynx.
In some embodiments, the signal processing system includes a slow envelope filter for providing a slow envelope signal. For example, the signal processing system can include an oscillator in communication with the slow envelope filter and with the voice actuator. The oscillator is configured to receive the slow envelope signal and to provide a pitch-control signal to the voice actuator. The pitch control signal depends, at least in part, on the slow envelope signal.
Or, the slow envelope filter can be a low-pass filters such filters can include those having a cutoff frequency of 1 Hz.
In other embodiments, the signal processing system includes a fast envelope filter for providing a fast envelope signal. Where a fast-envelope filter is provided, the voice prosthesis can include a switch in communication with the fast envelope filter and with the voice actuator. The switch is configured to actuate the voice actuator at least in part on the basis of the fast envelope signal. One suitable switch is a Schmitt trigger.
In other embodiments of the voice prosthesis, the signal processing system is configured to provide first and second control signals to the voice actuator. The first control signal controls a pitch of the voice actuator in response to the neural activity. The second control signal turns the voice actuator on and off in response to the neural activity.
Embodiments of the voice prosthesis include those in which the signal processing system is an analog signal processing system, as well as those in which the signal processing system is a digital signal processing system.
Another aspect of the invention includes a voice prosthesis having an electrode for detecting a measured signal indicative of neural activity, and first and second low-pass filters. The first low-pass filter has a first cutoff frequency, and an input configured to receive a signal derived from the measured signal. The second low-pass filter has a second cutoff frequency higher than that of the first low-pass filter. Its input is configured to receive the signal derived from the measured signals. An oscillator in communication with an output of the first low-pass filter is configured to provide a drive signal to a voice actuator. The drive signal has a drive frequency that depends in part on an output signal of the first low-pass filter. A switch in communication with an output of the second low-pass filter, is configured to provide an actuating signal to the voice actuator. The actuating signal depends at least in part on an output signal of the second low-pass filter.
In yet another aspect, the invention includes a method for controlling a voice prosthesis by detecting a signal indicative of neural activity; processing that detected signal to obtain a control signal representative of neural activity; and controlling a voice actuator of the voice prosthesis with that control signal.
Specific practices of the method include those in which detecting neural activity includes detecting activity associated with contraction of neck strap muscles, and those in which controlling a voice actuator includes controlling an electro-larynx.
These and other features of the invention will be apparent from the following detailed description and the accompanying figures, in which:
When a person attempts to speak, the brain causes nerve impulses to be transmitted to the laryngeal muscles. In a healthy person, these nerve impulses cause appropriate contractions of these muscles. To a great extent, a person modulates the pitch of his voice by modulating the signal that is transmitted to these laryngeal muscles.
A person who has had a laryngectomy has neither his larynx nor his laryngeal muscles available. However, in many cases, the neural signal that would otherwise control the laryngeal muscles or other neck strap muscles related to voice production remains available. The present invention harnesses these neural signals to control a laryngeal prosthesis.
Referring to
In one embodiment, the voice activator 16 is an electro-larynx having a diaphragm that is placed against the user's neck. A piston striking the diaphragm at a particular driving frequency generates vibrations that travel into the user's trachea. These vibrations enable the user to speak with a synthesized voice tuned to that driving frequency.
The transducer 12 is preferably a non-invasive transducer having an electrode that can be mounted on the surface of the skin. A particularly advantageous location for mounting the electrode is on the surface of the skin covering the neck strap muscles. With the electrode mounted at this location, EMG signals intended to control the laryngeal muscles can readily be measured.
The transducer 12 can also be an invasive transducer, such as a probe, that is placed nearer the source of the EMG signals. In addition, the probe or electrode can be placed at any location at which it can detect EMG or other neural signals intended to control the laryngeal muscles.
A suitable transducer 12 includes a bipolar differential electrode such as that provided by a DE2.1 manufactured by Del Sys Inc. of Boston Mass.
Referring now to
As a person speaks, the pitch of that person's voice changes. For example, a rising intonation is often associated with a question. In tonal languages, intonation can completely change the meaning of a phoneme. The frequency with which pitch is modulated in normal speech is, however, much lower than the frequencies that are present in the EMG signal. Even in tonal languages, it is unusual to modulate pitch more than once per spoken word.
If all the frequencies available in the EMG signal were used to modulate the pitch of the voice actuator 16, the result would be a voice having an unpleasant ululating quality, with a pitch that may change several times in the course of a single utterance. For this reason, the rectifier output signal is provided to a slow-envelope low-pass filter 24 (hereafter referred to as the “slow-envelope filter”) having a suitable cut-off frequency. In one embodiment, the slow-envelope filter 24 is a three-pole filter, such as that shown in
The cut-off frequency of the slow-envelope filter 24 is chosen to be low enough to avoid unpleasant pitch modulation, but high enough to avoid generating a monotonic, or robotic sounding voice. In one embodiment, a cut-off frequency (i.e., 3 dB corner frequency) of 1 Hertz has been found suitable for normal speech.
In the embodiment of
A pitch slope resistor R26 controls how rapidly the pitch changes with changes in the EMG signal amplitude. A suitable rate of change for normal speech is 112 Hz/volt.
Neither the pitch slope resistor R26 nor the pitch resistor R27 are changed by the user during normal operation. These values are typically adjusted once to customize the prosthesis to the user's neural and voice characteristics.
The slow envelope filter output (hereafter the “slow envelope”) is provided to an input of a voltage-controlled oscillator 26. The resulting output of the voltage-controlled oscillator 26 is a square wave having a fundamental frequency that depends on the amplitude of the slow envelope. This square wave is used to drive the voice actuator 16. In the case of the electro-larynx, the fundamental frequency of the square wave controls the drive frequency at which the piston strikes the diaphragm, and hence the pitch of the synthesized voice. The net effect is that the user controls the pitch of the synthesized voice in much the same way he would have controlled the pitch of his natural voice, namely by controlling the magnitude of the EMG signal of nerves normally used to trigger the laryngeal muscles.
The voltage-controlled oscillator 26 itself has settings to map input frequencies to output frequencies. These settings can be used to control the dynamic range of the synthesized voice. In the case of an analog VCO circuit, the VCO output frequency is a linear function of the input amplitude. However, by replacing the analog VCO circuit with a digital circuit having more complex digital signal processing capabilities, the mapping between input amplitude and output frequency can be altered. This would permit having an oscillator 26 whose output frequency is a non-linear function of the slow envelope amplitude.
Referring back to
In operation, the user excites nerves that would normally trigger either the laryngeal muscles or any neck-strap muscle used in connection with voice production. When the amplitude of the EMG signal generated by those nerves reaches an upper threshold value, the fast-envelope will have sufficient amplitude to cause the Schmitt trigger 30 to turn on the voice actuator 16. The voice actuator 16 will then produce an output at a frequency consistent with the output of the slow-envelope filter 24. When the EMG signal falls below a lower-threshold value, the output of the fast-envelope filter 28 causes the Schmitt trigger 30 to turn off the voice actuator 16.
The upper threshold value is selected to be low enough such that the user need not exert too much effort to turn on the voice actuator 16, but not so low that the voice actuator 16 is inadvertently turned on. The lower threshold value is selected to be low enough such that the voice actuator 16 does not turn off during low-pitch speech, but not so low that it becomes difficult to turn off the voice actuator 16.
The cutoff frequency of the fast-envelope filter 28 is selected to be low enough such that minor fluctuations in the EMG signal will not cause the voice actuator 16 to turn on and off, but not so low that the voice actuator 16 fails to turn on promptly at the onset of speech. In one embodiment, a cut-off frequency between 1 Hz and 16 Hz has been found to be suitable for many users.
While the embodiments described herein are implemented with analog circuits, it will be apparent that some or all the components can be implemented as digital circuits. Moreover, the components described herein can be incorporated into an integrated circuit, such as an ASIC.
Having described the invention, and a preferred embodiment thereof, what we claim as new and secured by letters patent is:
