Chapter 3 Basic Techniques for Speech & Audio Enhancement
Chapter 3 BASIC TECHNIQUES FOR AUDIO/SPEECH ENHANCEMENT 3.1 INTRODUCTION Audio/Speech signals have been essential for the verbal communication. They have found their significance in many commercial applications and even in entertainment industry and Interactive Voice Response System (IVRS). Various effects are rendered over audio and speech signals during the recording and editing of sound in entertainment industry. With the availability of advanced and efficient digital techniques and DSPs there have been a considerable growth in the number of effects which can be applied to the audio/speech signals and even the quality of the effects and ultimately the sound can be enhanced. Sound effects have a major role to play in music and entertainment industry, they are employed to make listening pleasurable and gripping. There are numerous commercial devices available which employ various techniques for the enhancement of audio signals by applying various effects like Echo, Reverb, Flanger, chorus, phaser and karaoke. The effects have become an integral part of Hi-Fi music systems, synthesizers, and gaming platforms and sound recorders. An equalizer is inevitable in music systems and sound recorders. The equalizers also aid in counteracting the Inter Symbol Interferences (ISI) in a communication channel which reduces the efficiency of a communication channel[rao]. This section of work evaluates some of the basic techniques for Audio and Speech signals using LabVIEW and implementing them on DSK C6713 DSP kit. The real time implementation of the effects has been achieved through the mydaq. The work is targeted towards the practical demonstration 41
of various Audio/Speech enhancement techniques in the classroom or laboratory instructions to undergraduate students. The techniques described in the present study will make students understand the concepts of Digital Signal Processing techniques like filtering for the enhancement of Audio/Speech signals on hand. 3.2 DIGITAL FILTERS A digital filter is a linear system that changes the amplitude or phase of one or more frequency components of an audio signal. Types of digital filter are FIR (finite-impulse response) filter IIR (infinite-impulse response) filter Digital filters effectively reduce the unwanted higher or lower order frequency components in a speech / Audio signal. The additive noise includes periodical noise, pulse noise, and broadband noise related problems. The noise generated by the engine is one kind of periodical noise while the one generated from explosion, bump, or discharge is pulse. There are many kinds of broadband noise, which may include heat noise, wind noise, quantization noise, and all kinds of random noise such as white noise and pink noise. Statistical relationship exists between the noise and speech; i.e. uncorrelated or even independent noise, and correlated noise (such as echo and reverberation). In acoustics applications, noise from the surrounding environment severely reduces the quality of speech and audio signals. Therefore, basic linear filters are used to de-noise the audio signals and enhance speech and audio signal quality. Our objective is of a noise reduction system with heavily dependent on the specific context and application as to increase the intelligibility or improve the overall speech perception quality which aimed to reduce unwanted ambient sound by implementing through different filters. 3.3 AUDIO / SPEECH NOISE REDUCTION SYSTEM Numerous researches and development has been done is current years to design a noise filter foraudio for the enhancement of audio signals. Though, there are several characteristics that need tobe deliberated before designing a 42
filter for the audio enhancement. In this section, we present a general idea of the characteristics involved and the filtering techniques used by each of thevarioussituation and circumstances. For cleaning audio / speech signal using digital filter system, one has to take certain precaution while selecting sampling rate as well as information about type of noise, intelligibility of signal and channel information. For said work, we took recorded speech using audacity software [41] with 1 KHz stationary noise. Noisy Speech signal is fed through AUDIO IN of TI DSK C6713 kit and filtered speech is taken out form AUDIO OUT of same kit. Processing and filtering part is done through LabVIEW which is connected through USB port of PC shown in figure 2.10. We tried to filter audio speech using various IIR, FIR (Low Pass, High Pass, Band Pass, Band Stop) filters using LabVIEW and also tried for phase reversal techniques for removing noise from audio speech and implemented on real time hardware DSP kit TI C6713. Developed VI removes noise from the audio signal. It is developed in LabVIEW. It is an effort to further grasp the fundamentals of LabVIEW and validate it as a powerful application tool. There are basically different files. Each of them consists of front panel and block diagram. These are the programmable files containing the information about the filter and figure files are the way to analyze the given audio and enter the various filter related data. In this work we had input speech in the format.wav. Listen the sound which will appear to be noisy. In the VI, option was there to choose the desired filter for de-noising which includes various parameters namely type, order, passband frequency, stop-band frequency, pass-band ripple and stop band ripple. Simultaneously front panel shows filtered sound. For this exercise, the sampling frequency of the input file is equivalent to the Nyquist frequency of 8000Hz. The bandlimit, or maximum frequency of the original signal, is 4000Hz. The maximum cutoff frequency must be 4000Hz as well. The VI was designed to filter the speech using various filters and display the frequency spectrum of the filtered signal. 43
Figure 3.1: Front Panel of Noise Reduction System using various Filter Figure 3.2: Block Diagram of Noise Reduction System using Various Filter Figure 3.3: Block Diagram of Noise Reduction System using Phase Shift Method 44
3.4 DIGITAL AUDIO EFFECTS In earlier days the audio effects were applied to the signals using the costly and bulky analog circuits [42] which used to occupy a lot of space and were prone to noise too. The advent of integrated circuits and advancement in digital systems the effects are applied by various digital components and circuits, which are relatively inexpensive, and low on size and weight. Even the quality of the applied effect is better compared to their analog counterparts. The audio effects are classified in three categories depending upon their techniques of production. The three categories being: Filter Effects, Delay Effects and Simulated Sound Effects. 3.4.1 FILTER EFFECTS The effects which can be applied on the signal by using digital filters enhance or alter them are categorised as Filter Effects. One of the most significant and common Filter effects Graphic Equalizer effect. A graphic equalizer employs several filters to allow different frequency bands to be manipulated separately and then adds them back together for output. Graphical equalizers are available with music systems and mp3 players where a user can select from preset gain options. 3.4.2 DELAY BASED EFFECTS The effects created by storing the incoming audio signal sample and then delaying it by accessing the stored signal sample later point in time are known as Delay based effects. Different methods to access delayed signal renders various delayed based effects such as reverberation, phasing, flanging, chorus, and echo effects. ECHO EFFECT Echo is a natural phenomenon based on the reflection of sound. When the sound waves produced by a source get reflected by a distant object they are reflected back. If the reflected sound is heard after 100 milliseconds or more of original sound, it can be heard distinctively and clearly. Then we can say that the sound is echoed. To produce echo effect the previously stored input 45
signal which is known as delayed signal is played back after a certain period of time along with the current input signal. This delayed signal may be played back either once for a singular echo or more than once for multiple echo. Thus echo is a replica of original audio signal but which has been delayed in time. In telecommunication echo signal is not desirable and needs to be removed or suppressed. But, an echo effect is widely used in music and entertainment industry to enhance the sound. A basic echo effect can be achieved by simply adding a previous sample to the present sample. Alternately, the simple echo effect requires two basic operations, namely, time delay and addition. The time delay may range between a few milliseconds to a few seconds. Figure 3.4: Representation of fundamental echo Basically, a delay time is provided to an input audio signal while it plays back, produces an echo. The delay time can range from some milliseconds to several seconds. Figure 3.4offers the fundamental echo in a flow-graph form. This is a single echo effect as it only produces a solitary copy of the applied audio signal. Figure 3.5: Representation of echo with feedback 46
Only having a single echo effect is some what restraining, so utmost delays also have a feedback control (occasionally called regeneration) which takes the output of the delay, and sends itback to the input, as shown in figure3.5. Now, you have the ability to repeat the sound over and over, and it becomes quieter each time it plays back (assuming that the feedback gain is less than one. Most delay devices restrict it to be less than one for stability). With the feedback, the sound is theoretically repeated forever, but after some point, it will becomeso low that it will be below the ambient noise in the system and inaudible. Many useful audio effects can be implemented using a delay structure. Basic delay structures out of some very basic FIR and IIR comb filters. Combination of FIR and IIR gives the Universal Comb Filter. The popular guitar effects can be implemented with a comb filter (FIR or IIR) and also some modulation effects. Flanger, Chorus, Slapback, Echo are same basic approach but different sound outputs as shown in Table 3.1 Table 3.1: Audio Effect delay settings Sr. No. Effect Delay Range (ms) Modulation 01 Resonator 0 20 None 02 Flanger 0 15 Sinusoidal 03 Chorus 10 25 Random 04 Echo 50 None IIR filter Simulates a single delay and endless reflections at both ends of cylinder output to input. The input signal circulates in delay line (delay time τ) that is fed back to the input. Each time it is fed back it is attenuated by gain. Input sometimes scaled by to compensate for high amplification of the structure. It s represented by the formula: ( )= The transfer function is: ( )+ ( ( ) = 1+ ) ℎ = Figure 3.6 and 3.7 representing delay effect with the help of IIR filter using LabVIEW. 47
Figure 3.6: Block Diagram of Delay Implementation using IIR special filter Figure 3.7: Front Panel of Delay Implementation using IIR special filter REVERB Sometimes the sound is still perceivable even if the source has stopped producing the sound. This phenomenon is known as reverberation. Prof. Sabine has carried out intense research about Reverberation in the field of acoustics. The time until which the sound is audible even after its source has stopped producing the sound is known as reverberation time. According to the definition of Reverberation given by Prof. Sabine After the source has ceased to produce the sound, the time taken to roll-off till the one millionth of its original intensity is known as Reverberation time. 48
There is a very thin boundary between the definitions of echo and reverberations and often thought upon as the same. Although one of the primary reasons behind the reverberation is the reflection or multiple reflections of sound it is not the same as echo. Circular buffer is one of the technique to generate reverberation and echo. A circular buffer works in a different way. Values within the buffer (array of data from audio signal) are not shuffled along. Instead, the oldest element is over-written by the newest. This means that the location of the oldest element in the buffer moves every time a new sample is saved. We keep track of the position of the oldest element with a pointer as shown in Figure 3.4, which assumes that the buffer already contains the values 1,2,3,4,5,6,7 and 8. Figure 3.8: Circular Buffer arrangement In order to insert a new value into the buffer, the oldest value is simply overwritten and the pointer moved along one place. Unlike a straight buffer, all the other values still remain in exactly the same place they are not shuffled along. When the end of the buffer is reached the oldest value is overwritten with the newest. Inthis case the pointer is set back to the beginning of the buffer again, as if the buffer werecircular. Implementation of reverberation in real time using LabVIEW and DSK TI C6713 is shown in Figure 3.9 and 3.10. 49
Figure 3.9: Block diagram of Reverberation Figure 3.10: Front Panel of Reverberation FLANGER EFFECT Flanger effect is an audio effect seen in many hi-fi audio systems and mixing consoles. Flanging means mixing of two similar audio signals (mostly the same audio signal). One of the signal being delayed by very small duration and varying delay time. The resultant output is similar to the output of a swept comb FIR filter, with peaks and notches of the harmonics spread up and down in the output spectrum. Figure 3.11 and 3.12 represents the Flanger effect using filter. 50
Figure 3.11: Block diagram of Flanger Effect Figure 3.12: Front Panel of Flanger Effect 3.4.3 SIMULATED SOUND EFFECTS The effects produced by digital signal processor/microprocessor or any other programmable device to generate digital form of sounds are known as simulated sound effects. These types of sound are generated by various algorithms and filters. These techniques facilitate creation of sounds mostly heard in techno music as well as in digital keyboards for synthesis of various sounds along with its own music. The real time synthesis of audio effects are implemented using LabVIEW on DSK TI C6713kit as shown in Figure 3.13 and 3.14. 51
Figure 3.13: Block Diagram of Synthesis Audio Figure 3.14: Front Panel of Synthesise Audio 52
3.5 GRAPHIC EQUALIZER Equalizer is a system capable of attenuating and boosting the signal in different frequency bands as desired [43]. Graphical equalizers used widely in consumer and high end professional audio devices and systems. They are comprised of a bank of sliders for boosting or attenuating different frequencies. The name graphic equalizer has been given because they give the ability to the user to graphically draw the frequency attenuation using a set of sliders. Typicalattenuation range of agraphic equalizers is from 12 db to +12 db. Graphic Equalizers are available in 10-band (1 octave), 15-band (2/3 octave), or 31-band (1/3 octave) formats. The center frequencies of 10-band equalizer filters are an ISO standard, and are given in Table 3.2. Band Frequency 1 2 3 4 5 6 7 8 9 10 31.5 63 125 250 500 1 2 4 8 16 Hz Hz Hz Hz Hz KHz KHz KHz KHz KHz Table 3.2: Frequency bands of a 10-band Equalizer The input signal is fed into each of the 10 filters, every filter extracts an individual frequency band which is attenuated by a variable gain. These attenuated signals are then reconstructed by summing them together. A low pass and a high pass filter are used for the lowest highest frequency band respectively. The pass band of each filter is so designed that it is 33% of the distance between center frequencies. Hence, 33% of the area between two filters fallsunder transition bands. As the transition bands overlap each other, the gain of the two neighbouring filters will affect the overlapping area. Figure 3.15: System implementing music equalizer using TMS320C6713 53
The system in Figure 3.15 is an audio equalizer that typically adjusts the energy levels of the audio data in one or moredifferent frequency bands in order to change the characteristics of the audio data. This equalizer allows boosting (or suppression/attenuation) of frequencies between the source of a sound and the output of the sound. The setup for the 10 band Graphic Equalizer consists of thehardware and software where we need TMS320C6713 DSP kit, Personal Computer (PC) with LabVIEW, a universal synchronous bus (USB) cable which connects the DSK and the PC, a 5V power supply to connect to DSK, Oscillator, Signal Generator and Speaker [5] as shown in the Figure3.16 Figure 3.16: Graphic Equalizer Experimental Setup The front panel of the task is developed on LabVIEW as in Figure 3.13. The block diagram description the new templateis opened by selecting the option from windows menu. Inthe project three stacked sequence structured block diagram where each stack contains one or more sub diagrams that executes sequentially as shown in Figure 3.17. The necessary functions were added in the stacks to create the objects in the front panel. The LabVIEW block diagram designed for the Graphic Equalizer. 54
The audio input is applied to Line-in of DSK which is taken either from a PC or through a mobile. Mono channelis used for simplicity. The sampling frequency is set to 48kHz. Word length is 16-bit, scaling is normalized and samples per frame are 64. Simulink model for equalization are built with the help of Simulink block sets and its equivalent C code is transferred to the DSP Starter Kitwith the help of USB cable. Actual signal processing is done in DSK with the help of C code downloaded in it which generates an equalized music output. Output is takenfrom the Headphone out of the DSK and is applied to speaker. Figure 3.17: Block diagram and Front Panel of 10 Band Graphic Equalizer 55
In this way, the design and implementation of 10 filters for each center frequency is done and at last outputs of these filters is added and given to the output sample where the equalized controllable output is obtained from the speaker. 3.6 VIRTUAL KARAOKE MACHINE Karaoke word is derived from two Japanese words Kara which means empty and Okesutora which means orchestra. Karaoke is a form of entertainment in which users sing along with recorded music using a microphone [44]. Dedicated Karaoke machines came into existence since early 1970s and later on the Karaoke feature was incorporate in commercial music players. Karaoke machines remove the original voice of the artist from a song. If the Suppression of voice is done by equalizing the original signal for attenuating the vocal frequencies, we have to use several filters of different pass band frequencies. But filters cannot suppressthe vocal band from the signal completely and affects the original signal by changing or distorting the pitch of vocal band, suppressing and distorting the bass component and ultimately changing the spectral composition of the original signal. Figure 3.18: Karaoke System Design 56
Recording of stereo sound where there are two channels is done in a way where the vocal band components are same in both the channels and the components of other musical instruments are different in them. In order to reduce the vocal band components, the stereo sound signal is split in two components. As a result two mono sound tracks (left and right) are obtained. Thereafter one channel is subtracted from other channel. Due to this voice of the original artist is considerably reduced and can be suppressed completely if the recorded music is perfectly balanced. Figure 3.18 represents the method of original voice suppression with the help of Karaoke Machine. Figure 3.19: Karaoke System Implementation in real time Karaoke Machine was implemented using LabVIEW and TI DSKC6713 kit as shown in figure 3.19. In LabView VI, audio file is read and then separated into two parts-left and right channel. Left Channel is multiplied using positive value of 1 and same way right channel is multiplied with -1 integer value. Multiplication had altered right channel. After that both the channels are added. This results in the removal of same component and presence of different component in the output. 57
Figure 3.20: Block diagram and Front Panel of Karaoke System using LabVIEW 58
3.7 RESULTS AND DISCUSSION In the present study various filters with different frequencies are used to de-noise the audio and speech signals. Faithfull results of de-noising have been obtained for low level noise. Although it has been observed that we cannot remove the noise completely without affecting the audio signal, but with the perspective of making students understand the concepts of various filters the results obtained are encouraging. We have also studied the technique of noise suppression by the phase reversal technique, where upon considerably good results have been observed for the removal of noise from signal of interest. In the next section we have tried to implement various audio effects such as Echo, Reverb, Flanger, Simulated Sounds and Graphic Equalizer. The Echo and Reverb effects having various delay times have been observed and successfully implemented on the DSK6713. Similarly the implementation of Flanger effect where the time delay is constantly varying has been analyzed and implemented on DSK6713. The Graphic Equalizer which has been designed using LabVIEW and again implemented on TMS320C6713 DSP is found to enhance the quality of audio signals by adjusting the gains for different frequency bands. The virtual karaoke machine implemented in the present study focuses on the approach to remove the voice signal from the audio. We have provided the system balance both the channels, left and right by adjusting the gains and also provided the volume control to boost the output audio signal Thus overall this chapter of the present work discussed the aspects of enabling the student grasp basic concepts of Digital Signal Processing and learn basic programming skills related to DSP using LabVIEW. This section also enables the students to understand the technique to implement the basic techniques of DSP over the hardware such as DSK6713 for their real time applications. 59