WEARABLE SPEECH ENHANCEMENT

Size: px

Start display at page:

Download "WEARABLE SPEECH ENHANCEMENT"

Brett Long
5 years ago
Views:

Proceedings of the 2004/2005 Spring Multi-Disciplinary Engineering Design Conference Kate Gleason College of Engineering Rochester Institute of Technology Rochester, New York 14623 May 13, 2005

1 Proceedings of the 2004/2005 Spring Multi-Disciplinary Engineering Design Conference Kate Gleason College of Engineering Rochester Institute of Technology Rochester, New York May 13, 2005 Project Number: WEARABLE SPEECH ENHANCEMENT Carl Audet John Dimmick Brandon Mikulis ABSTRACT This paper outlines the design of a speech enhancement device for Parkinson s patients. The device amplifies speech while blocking ambient noise and acoustic feedback through the use of a high-tech acoustic sensor that couples directly to the skin. Additionally, bandpass filtering and compression amplification allow the output to be noise-free with minimal distortion. FM transmission is used to transfer the signal from the acoustic sensor to the speaker. Lastly, low power op-amps and a shutdown mode are used to improve battery life. The final design has been prototyped and has a positive outlook for production. attempts Dr. Perlman has provided a possible solution to these problems through the use of a new acoustic sensor. NOMENCLATURE ARL Army Research Laboratory FCC Federal Communications Commission SNR Signal-to-Noise Ratio V p-p Peak to Peak Voltage INTRODUCTION Patients with Parkinson s disease often experience speech impairments. One such impairment is the loss of voice intensity. As a result, these patients are difficult to understand in a noisy environment. This loss in voice intensity can leave the patient and the people who they come in contact with frustrated over their inability to effectively communicate. It may be possible to overcome this issue through the use of voice amplification. The objective of this project is to create a device that will amplify the voice of a Parkinson s patient to the point that it can be easily heard and understood in a noisy environment. In previous attempts a microphone has been attached to a patient s shirt collar or placed in close vicinity of their mouth. Unfortunately, this configuration allows background noise and acoustic feedback from the speaker to be amplified with the signal. Based on these previous REQUIREMENTS The wearable speech enhancement device will output intelligible speech at a level that can be easily heard at a close range in a noisy environment. The output will limit noise and be free of acoustic feedback. This device is to be minimally intrusive in the user s daily life and for this reason it will be lightweight and portable. Additionally, the acoustic sensor must be attached in a manner that is both comfortable and secure. In order to provide greater portability the system is split into two parts, 1) the sensor-transmitter and 2) the receiver-output. Since both sections are portable they need to be battery operated. Separation of the acoustic sensor and the output speaker enables greater usability in diverse situations. The user can have the output near them if they are in a 2005 Rochester Institute of Technology

2 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 2 conversation with another individual in a quiet room. If the user is in a crowded situation or a loud environment the output speaker can be placed near the person or persons they are talking too. As in normal social situations one does not carry conversations with people on the opposite side of the room the transmitter-receiver link only needs to operate over a range of a few feet. SYSTEM OVERVIEW The system is broken down into two modules. In the first module the speech signal is acquired through the acoustic sensor and then amplified and transmitted. Amplification is preformed by the pre-amp that comes with the ARL sensor. The amplification occurs in two stages; a instrumentation amplifier followed by an active bandpass filter with large pass band and a gain. Figure 1 shows the block diagram of the first module of the device. Fig. 1: Block diagram of the first module. In the second module the signal is received and then filtered and amplified before it is output to the speaker. Directly after the signal is received it is buffered and then goes through an active bandpass filter with a passband similar to the bandwidth of the telephone system. Next, the signal undergoes dynamic range compression and amplification by a compression amplifier. Any additional high frequency noise that is accumulated in this stage is attenuated by a passive low-pass filter. Finally the signal is amplified by an audio power amplifier with user-controlled gain and output to the speaker. Figure 2 shows the block diagram of the second module. Fig. 2: Block diagram of the second module. ACOUSTIC SENSOR ratio. Likewise, it can be used to pick up a person s voice by placing it on the throat. The sensor is an active acoustic sensor that uses a liquid filled bladder to physically couple to the user s skin. The bladder and liquid contained within were chosen to have similar physical properties to skin, thereby minimizing the loss of vocal signals due to unmatched acoustical impedance. Additionally, the physical coupling helps to block ambient noise and acoustic feedback. A major setback of the ARL sensor is the cost, as it is a technology still in development, but was offset by the sponsor purchasing it for the project. Licenses to produce the sensor are available to purchase if the sensor is to be used in a production product. PRE-AMPLIFIER A pre-amplifier is required to amplify the signal so that it can be transmitted with a high SNR. The signal coming from the ARL sensor is on the order of tens of millivolts. The transmitter can take input voltages from 0 V to 3 V p-p. In order to receive the signal with high SNR, it should be amplified as much as possible (without creating distortion) before it is transmitted. The ARL sensor came with a preamplifier attached. The circuit operates on a 3 volt supply and draws only a few milliamps of current. The design is split into three distinct stages as shown by figure 3. Fig.3: ARL pre-amplifier circuit. The first stage is a simple voltage divider that sets a DC voltage to power the sensor and act as a DC offset for the signal. The second stage is an instrumentation amplifier with a fixed gain of about 2. The final stage is an active bandpass filter with a passband from 2.5 Hz to 9.5 khz and variable gain. This frequency range easily encompasses speech frequencies and will not distort the signal. The preamp was simulated using Capture to verify its frequency response and gain. The results of that simulation are shown by figure 4. Previous attempts at this project have lead Dr. Perlman to an acoustic sensor that is currently being developed by the U.S. Army Research Laboratory. The sensor was developed to acquire biological signals from officers in the field with a high signal-to-noise Paper Number 05201

3 Proceedings of the Winter KGCOE Multi-Disciplinary Engineering Design Conference Page 3 FILTERING Fig. 4: ARL preamp frequency response. Since the ARL preamp meets the system requirements and it is readily available, it is included in the final design. However, during the redesign for board layout the on/off switch was removed and the gain in the active filtering fixed at about 13. WIRELESS TRANSMISSION In order to improve the functionality and comfort of the device, wireless transmission is used to relay the signal from the preamp to the filters, amplifiers and speaker. The wireless link consists of an FM transmitter (TXM-916-ES) and receiver (RXM-916- ES) pair produced by Linx Technologies. This transmitter-receiver pair was chosen because its modulation frequency of MHz falls within the band from 902 MHz to 928 MHz designated by the FCC for continuous audio transmission without an operating license [1]. The transmitter specifications state the transmitter is capable of transmitting up to meters [2], which is more than is required for this project. The transmission distance can be decreased by decreasing the transmission power. This is done by either using an inefficient antenna or adjusting the transmission power on the TXM-916-ES using an external resister and the level adjust pin (LADJ). The maximum output power into a 50 load is +4dBm and is adjustable over a range of about 65dBm [2]. This output power is set using a fixed resister that will be inaccessible to the user in compliance with FCC regulations. This resistor value would be determined during qualification testing for FCC regulations if it is to become a consumer product. Another beneficial feature of this transmitter and receiver pair is their ease of implementation. They require no external RF components besides antennas. These antennas will be contained within the housing of the speech enhancement device making them inaccessible by the user for modification in compliance with FCC regulations. The transmitter specifications state that it accepts a 0-3 V p-p analog signal. However, the best results were found using a 1 V p-p signal. The small signal from the ARL acoustic sensor and low gain of the preamp ensures that this voltage level is not exceeded. Through feasibility assessments and advisor meetings, it was determined that digital signal processing would be excessive for this design, as a result analog filtering techniques were chosen. The main components of speech are contained in a frequency band from 300Hz to 3.2kHz. A Chebyshev design approximation was used to implement an active bandpass filter with a passband of 100Hz to 3.2kHz; this was found to produce the best sounding speech. The Chebyshev method was chosen because it combines a low order design with a steep increase in attenuation [3]. Two Delyiannis-Friend stages make up the circuit and provide 40 db/decade attenuation at each pole. Figure 5 shows the bandpass filter circuit while figure 6 shows the simulation of the frequency response of the filter. R1 9.91k R6 100k C1.1u C2.1u 0 0 R5 26k - + U1 OPAMP OUT R2 3.3k R3 180k C4.01u C5.01u 0 0 Fig. 5: Bandpass filter circuit. Fig. 6: Bandpass filter frequency response. It can also be seen from Fig. 6 that there is 20 db of attenuation in the passband. This serves to decrease the noise amplitude so that more of the noise riding on the signal will be eliminated by the compression amplifier that follows. Since the noise gate threshold of the compression amplifier is 500 V, all signals bellow that level will not be amplified. As a result, the compression amplifier that follows will greatly improve the SNR. The total power consumed by the 2 nd order bandpass filter is approximately 8 mw. After the compression amplifier additional filtering is performed by a basic R-C low-pass filter to remove any additional high-frequency noise that was amplified by the compression amplifier. This filter greatly improves the speech quality. R4 9.8k - + U2 OPAMP OUT Copyright 2005 by Rochester Institute of Technology

4 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 4 COMPRESION AMPLIFIER At times the amplitude of the user s voice may fluctuate and become very loud or quiet. In order to produce normal speech these voice fluctuations need to be suppressed. Unfortunately, they cannot be completely eliminated since they arise in normal speech patterns, however the range of the fluctuations can be limited. This can be done through dynamic range compression. Dynamic range compression amplifies the louder parts of the signal less than the quieter parts, thus compressing the range. The change in amplification in relation to input intensity is logarithmic. The SSM2165 by Analog Devices was chosen to perform dynamic range compression in this device. This part provides variable compression along with noise gating and a limiting threshold. The variable compression is controlled by an external resister which allows for a compression ratio between 1:1 to 15:1 [4]. The noise gating improves SNR by suppressing signals below 500µV, which in this case is noise. The limiting threshold is the point at which the compression amplifier operates with a fixed compression ratio of 10:1, limiting the loudest signals. AUDIO POWER AMPLIFIER The final amplifier is an LM4871 audio power amplifier from National Semiconductor that is designed to drive an 8 ohm load. The amplifier circuit is very easily implemented and eliminates the issue of impedance matching. It is mono bridged and capable of supplying 3 W of continuous power with about 0.1 % total harmonic distortion and noise [5]. It operates on a 5 V supply and draws only 6.5 ma of DC current. The LM4871 also has an external gain setting that serves as the master volume control. This allows the user to adjust the gain by up to a factor of ten. virtual ground is used to reference the signals and enable a single supply operation of the opamps. LOW POWER MODE Since the wearable speech enhancement device is designed to be a battery powered device, it is necessary to limit power consumption. In order to increase battery life low-power opamps were selected for implementation of the active filter. Additionally, a power save mode was created to turn off the audio power amplifier and the active filters when the user is not talking, or when the transmitter is off or out of range. The RXM-916-ES receiver provides two output signals useful for generating a power save mode; a received signal strength indicator (RSSI) and an audio reference (AREF) voltage. Both of these signals are DC voltages that change to reflect signal strength and audio level. These, in combination with a simple comparator circuit provide the control signal to remove the power from the audio amplifier and active filters. In this scheme the control signal is used to switch a transistor located between the +5V rail and the V dd pins of the opamps and audio amplifier. A diode is reverse biased across the load to act as a simple snubber circuit. The snubber circuit protects the transistor from the inductive properties of the load once the switch is opened. Figure 7 contains the comparator circuit and the switch with a snubber circuit. POWER SYSTEM Two separate power systems are utilized in the wearable speech enhancement device and both operate off of batteries. The first system is powered by a single 9V battery in series with a 3V voltage regulator to provide a steady supply for the ARL sensor, preamp, and the transmitter. The second system is powered by four AA (1.5V) batteries in series with 5V and 2.5V regulators providing the voltages to the individual components. The 5V regulator provides the supply voltage for the receiver, opamps and amplifiers. All of the components are referenced to the return terminal of the batteries. However, since audio signals swing both positive and negative, a 2.5V Fig. 7: Comparator and transistor switch with protection. Paper Number 05201

5 Proceedings of the Winter KGCOE Multi-Disciplinary Engineering Design Conference Page 5 BATTERY LIFE The transmitter portion of the wearable speech enhancement system is powered by a 9V battery. Though testing the transmitter module it has been determined to draw 9mA of current. At a constant current draw of 9mA, a 9V battery will reach a cutoff voltage of 4.8V after 60 hours [6]. The circuitry contained in the transmitter module of the wearable speech enhancement system operates on 3V. Therefore the transmitter module is capable of at least 60 hours of continuous operation. The receiver portion of the wearable speech enhancement system is powered by 4 AA batteries. Through testing the receiver module has been determined to draw approximately 25mA of current. The AA batteries will suffer a voltage drop of.25v each after 45 hours [7] of continuous operation under these conditions. After this time the batteries will no longer be able to supply the circuitry in the receiver module with the 5V required for operation. HOUSINGS The ARL acoustic sensor comes in a black plastic case that is secured to the user s neck by a black elastic strap to ensure a secure connection is maintained between the patient s voice box and the sensor. The ARL sensor is connected via a cable to the preamp and transmitter housing. This enclosure is 4.2 x 2.5 x 0.85 and comes with a pocket clip for easy attachment. It contains a built-in compartment for a 9V battery, which makes replacement easy for the user. The enclosure also provides a user interface consisting of an on/off switch and a power indicator LED. The receiver, filters, amplifiers and speaker are all contained in another housing that is 4.72 x 4.72 x A user interface consisting of an on/off switch, volume control knob, and a power indicator LED is mounted on the top of the enclosure. The batteries are easily accessible from the back of the enclosure. This enclosure can be attached to a wheelchair through the use of two Velcro straps or carried. IMPLEMENTATION Implementation of the wearable speech enhancement device was neither straight forward nor simple. The largest hurdle was overcoming noise. Both amplification and wireless transmission are large sources of noise. To cut down on noise in amplification and active filtering stages a large ground plane was used with numerous connections and filtering capacitors. Given that the FM transmitter and receiver were self contained chips, the only external components used were antennas and filtering capacitors. To maximize signal transmission, quarter wavelength whip antennas that matched the impedance (50 ) of the antenna pins were used with the transmitter and receiver. Greater signal strength will be achieved when the circuits are built on a printed circuit board as the prototyping perf board is poor for the performance of transmitters and receivers operating in the 900MHz range. CONCLUSION The results show that a portable speech enhancement device can be constructed to limit acoustic feedback and background noise. This design provides proof of concept and leads the way to an original prototype incorporating all of the desired features. ACKNOWLEDGEMENTS We would like to thank the following people for their help on this project: Dr Perlman, Dr Amuso, Dr Philips, Professor Garsin, and the RIT Electrical Engineering Department. REFERENCES [1] Federal Communications Commission, Code of Federal Regulations Title 47 part 15: Radio frequency Devices, [Online Document] FCC, June 16, Available at HTTP: [2] Linx Technologies Inc., ES-Series Transmitter Manual [Online Document] Linx Technologies Inc Available at HTTP: ducts_cat/rf_modules/es_series/estxm_manual.pdf [3] R. Schaumann and M. E. Van Valkenburg, Design of Analog Filters, England: Oxford Press, [4] Analog Devices Inc., SSM2165_b [Online Document] Analog Devices Inc Available at HTTP: Sheets/ SSM2165_b.pdf [5] National Semiconductor Corporation, LM4871 [Online Document] National Semiconductor Corporation, January Available at HTTP: Copyright 2005 by Rochester Institute of Technology

6 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 6 [6] Energizer Holdings Inc., 522 [Online Document] Eveready Battery Company Available at HTTP: [7] Energizer Holdings Inc., e91 [Online Document] Eveready Battery Company Available at HTTP: Paper Number 05201

Receiver Design. Prof. Tzong-Lin Wu EMC Laboratory Department of Electrical Engineering National Taiwan University 2011/2/21

Receiver Design. Prof. Tzong-Lin Wu EMC Laboratory Department of Electrical Engineering National Taiwan University 2011/2/21 Receiver Design Prof. Tzong-Lin Wu EMC Laboratory Department of Electrical Engineering National Taiwan University 2011/2/21 MW & RF Design / Prof. T. -L. Wu 1 The receiver mush be very sensitive to -110dBm