Cancel
Showing results for 
Search instead for 
Did you mean: 

Articulation Index

Siemens Phenom Siemens Phenom
Siemens Phenom

(view in My Videos)

 

Articulation Index

 

Articulation Index (AI) is a sound metric that was developed to indicate how much background sound levels can interfere with human speech.

 

It has a value that ranges between 0% (no speech understood) to 100% (all speech understood).  

 

Originally, Articulation Index was used in applications like measuring speech privacy and the effectiveness of communication systems.  Today, Articulation Index is also used to rate vehicle interior noise, the quietness of white goods, and more.

 

This article has the following sections:

  1. Background

       1.1 Human Speech Frequency Range

       1.2 Masking and Octave Bands

  2. Articulation Index Calculation

        2.1 Perform Background Sound Measurement

        2.2 A-weighted octave spectrum

        2.3 Determine Octave Band Coverage

        2.4 Multiply Coverage by Weightings

  3. Articulation Index Example Calculations

        3.1: Zero Percent Articulation Index

        3.2: One Hundred Percent Articulation Index

        3.3: Fifty Percent Articulation Index

        3.4: Ninety Nine Percent Articulation Index

        3.5: Forty Three Percent Articulation Index

  4. Articulation Index and Speech Intelligibility

  5. Articulation Index Applications

         5.1 Warning Chime

         5.2 Interior Acoustics

  6. Modified or Open Articulation Index

  7. Calculating Articulation Index in Testlab

  8. Calculating Articulation Index in Simcenter Testlab Neo

        8.1 Getting Started with Simcenter Testlab Neo Process Designer

        8.2 Using Simcenter Testlab Neo Process Designer

 

1. Background

 

Two key human hearing phenomena influence the Articulation Index: human speech frequency range and masking and octave bands:

 

1.1 Human Speech Frequency Range

 

Even though the humans can hear from 20 Hertz to 20,000 Hertz, the frequencies produced in human speech cover a much narrower range. In Figure 1, the light blue area highlights the frequency and amplitudes that a human can hear.  The orange area indicates the frequencies and amplitudes produced from normal human speech.

 

Human_Hearing_Map.pngFigure 1: Map of human hearing audio range with sound level in decibels versus frequency (light blue). The human speech frequency range (orange), is critical to speech being understood properly. Any background sound that occurs in this frequency range interferes with human speech and makes it more difficult to understand.

The speech range covers from 200 Hertz to about 6000 Hertz. When Articulation Index is calculated, the background sound levels that occur within this frequency range are given the highest importance, since they will interfere with human speech.  Background sounds outside of this range are not important.

 

1.2 Masking and Octave Bands

 

If the level of background noise is high enough, it will mask other sounds, including human speech. Masking means that it makes the speech inaudible: in other words it covers it up, or blocks it.

 

To understand the masking potential of a background sound, it is broken into individual frequency ranges that mimic human hearing.  The human ear, due to the construction of the hearing organ (called the cochlea), tends to aggregate sound into these frequency bands. These bands are called octave bands

 

The sound levels of one third octave bands covering from 200 Hz to 6300 Hz are used to calculate Articulation Index. These bands become wider and wider with increasing frequency as shown in Figure 2.

 

OctaveBands_LinearFrequencyScale.pngFigure 2: Sound spectrum in octave bands on a linear scale. Bands are noticeable wider with increasing frequency.

By convention, the octave bands are shown in equal widths on an octave scale as shown in Figure 3. The higher frequency bands cover a wider frequency range, but this is not obvious in the graph due to the frequency scale being in octaves rather than linear frequency.

 

OctaveBands_OctaveFrequencyScale.pngFigure 3: Sound spectrum in octave bands on an octave scale. Bands appear to be the same width with increasing frequency.

For more information on Masking and Octave, see the following Knowledge Base articles:

 

2. Articulation Index Calculation

 

One of the first definitions of Articulation Index was developed by Leo Beranek at Harvard University during World War II.  He used it to rate the effectiveness of various headsets used in airplanes. The calculation was outlined in "The design of speech communication systems", Proceedings of the IRE, Vol 45, 880-884, 1947.  It is also defined in ANSI S3.5-1969 and ANSI S3.5-1997.

 

While there can be slight variations in each standard, the following steps are typically performed in the calculation of Articulation Index:

 

2.1 Perform Background Sound Measurement

 

Perform a measurement of the background sound at the location where a listener is positioned. For example, the interior of a vehicle at cruising speed, the inside of a cockpit during flight, etc. Articulation Index is typically used on broadband background sounds.  Broadband sounds have a wide frequency range (think of the roar of a waterfall) with minimal tonal content.

 

2.2 A-weighted octave spectrum

 

Calculate an A-weighted octave spectrum in decibels from the measurement performed in the previous step.

 

2.3 Determine Octave Band Coverage

 

Plot the background sound against the chart in Figure 4 to determine how much the articulation window of each 1/3rd octave band is "covered". For example, if the sound level goes above the articulation upper limit (red line in Figure 4), the window of the octave band is completely "covered".  If the level is between the upper and lower limits, the octave band is partially "covered".

 

ArticulationWindow.pngFigure 4: The Articulation window is defined by a lower and upper limit, separated by 30 dB, for each octave band. If all octaves of background sound are below the lower limit, 100% of all speech is understood. If all octaves of background sound are above the upper limit, 0% of speech is understood.

A coverage value for each 1/3 octave band is calculated.  The window for each 1/3 octave band is 30 dB wide between the lower and upper limits:

  • If value falls above the upper limit, the result is 0 for that particular 1/3 octave band. The entire articulation window is covered.
  • If value falls below the lower limit, the result is 1 for that particular 1/3 octave band. None of the articulation window is covered.
  • If the A-weighted 1/3 octave level lies between upper and lower limits, then the value will be between 1 and 0.  It is calculated linearly between the upper and lower limit.  For example, if the background sound level for a particular octave band was 10 dB above the lower limit, the coverage value for that band is 0.666 or two thirds.

 

The exact values that define the upper and lower limits of the articulation window are defined in Figure 5. In the next step of the calculation, the coverage values are weighted.

 

2.4 Multiply Coverage by Weightings

 

The coverage values from the previous step are then multiplied by the weighting factors listed in Figure 5 below.

 

Weightings.pngFigure 5: Articulation Index weighting factors for octaves from 200 Hz to 6300 Hz. Octave bands most pertinent to human speech are weighted the highest.

Notice the weightings place more importance on octave bands that are critical to human speech.  Even though human speech occurs in octave bands ranging from 200 Hz to 6300 Hz, the most heavily weighted octave bands range from 1000 Hz to 4000 Hz. These weighting factors sum up to 100, which is the highest value for Articulation Index. In percentage terms, this is 100%.

 

After weighting is applied, the weighted coverage values for each octave band are summed to produce a single Articulation Index number. 

 

Some example calculations are detailed in the next section.

 

3. Articulation Index Example Calculations

 

Examples of measured octave data and their corresponding Articulation Index values are detailed below.

 

3.1: Zero Percent Articulation Index

 

If the levels of all octaves exceed the upper band limit (in red), no speech can be understood  (see Figure 6). 

 

ZeroAI.pngFigure 6: Background sound levels (blue bars) at the listener’s ear are above the upper limit (red). 0% of what is being spoken can be understood.

In this case, the background levels are so high, they interfere completely with speech and cover the articulation window of each octave band completely.

 

Every weighting value is multiplied by a value of zero (since each octave band’s articulation window is complete covered), so the sum of the weighted values is zero. This is an Articulation Index of 0%.

 

3.2: One Hundred Percent Articulation Index

 

If all decibel levels are under the lower limit (in green), there is no interference with speech (see Figure 7).

 

OneHundredAI.pngFigure 7: Background sound levels (green bars) at the listener’s ear are below the lower limit (green). 100% of what is being spoken can be understood.

 

All of the octave band coverage values are equal to one.  After multiplying the weighting factors by one, and summing, the total is 100.  In percentage, this is 100%.  The levels are too low to interfere with human speech.

 

3.3: Fifty Percent Articulation Index

 

If the levels cover the area between the upper and lower limits, the Articulation Index will be between 0 and 100 percent.  For example, in Figure 8, this is 50%.

 

FiftyAI.pngFigure 8: Half of the intelligibility window is covered, resulting in an Articulation Index of 50%.

Because half of each octave band is covered, the coverage values are all 0.5.  After multiplying the weighting factors by 0.5, the Articulation Index sum is 50, or 50%.

 

3.4: Ninety Nine Percent Articulation Index

 

In Figure 9 below, the sound spectrum has high levels in the 200 Hertz octave band that completely cover the intelligibility window.  The coverage value of this octave band is zero, while all other octave bands are one.

 

NinetyNinePercentAI.pngFigure 9: When the 200 Hertz octave band is completely covered, the Articulation Index is 99%. The 200 Hertz octave band does not interfere significantly with human speech.

The 200 Hertz octave band can deduct 1% maximum from the Articulation Index due to the weighting factors listed in Figure 5.  The 200 Hertz band has a weighting factor of 1, which is the lowest weighting compared to other octave bands. The coverage value for the 200 Hz octave band is zero.  So the weighted coverage value for this band is zero.

 

All the other bands have coverage values of one. When the weighted coverage values for the other bands are summed up, the total is 99.  The Articulation Index is 99% in this case.  There will not be much difficulty understand human speech with this background noise level.

 

3.5: Forty Three Percent Articulation Index

 

If six out of the sixteen octave bands, from 1000 Hz to 3150 Hz, are completely covered, then the Articulation Index is ~43%. (Figure 10).

 

FortyThreePercentAI.pngFigure 10: When the six bands, from 1000 to 3150 Hertz, completely cover the intelligibility window, the Articulation Index is 43%.

Each band between 1000 Hertz and 3150 Hertz would have a value of zero. All other bands outside of this range, have a value of one.  When weighted and summed together, the other bands have a total of 43, for an Articulation Index of 43%.  The bands between 1000 Hertz and 3150 Hertz were responsible for removing 57% from the total.

 

This underscores the importance of sensitivity of the human ear to frequency, as captured by the weighting values.  In this case, less than half of the octave bands are covered with high sound levels (6 out of 16 total bands), yet the Articulation index is less than 50% because those bands are critical to understanding human speech.

 

4. Articulation Index and Speech Intelligibility

 

Articulation Index can also be used to predict the percentage of syllables that can be understood.  This not only depends on the background sound levels (as measured by the articulation index) but on the speech itself.  

 

If the speech consists of sentences known to the listener in advance, a higher percentage will be understood. Speech consisting of random syllables and sounds have a much lower amount that will be understood.  This is shown in Figure 11.

 

AI_SI.pngFigure 11: For different types of speech, relationship between Percentage of Syllables Understood versus Articulation Index.

The relationship between the Percentage of Syllables Understood and the Articulation Index Percentage is not linear.  

 

5. Articulation Index Applications

 

Application examples using Articulation Index are described in this section.

 

5.1 Warning Chime

 

A manufacturer is considering different warning chimes (Figure 12) to alert the operator of the product about a potential problem. In this case, the warning chime sound with the lowest articulation index is preferred. 

 

chime.jpgFigure 13: Warning chimes need to interfere with human speech to be effective.

The alarm must interfere with human speech to gain the operator’s attention.  It would not be good if the operator was engaged in conversation and did not hear a warning.  For example, if the transmission failed while cruising at highway speeds.

 

5.2 Interior Acoustics

 

For occupants of vehicles like cars, trucks, bulldozers, and airplanes, a high articulation index is often used as a target, to ensure a quiet interior for the operator and occupants of the vehicle.

 

For example, in most modern passenger cars and sport utility vehicles, an Articulation Index of 70% or higher is targeted over the operating range of the vehicle.

 

In Figure 13, the Articulation Index of a vehicle is shown versus engine rpm under wide open throttle and partial throttle conditions.  This type of test is typically performed in 3rd gear of the vehicle.

 

AI_POT_WOT.pngFigure 13: Articulation Index versus Engine RPM at the operator ear location for a motor vehicle. Green trace is for a wide open throttle condition, while the blue trace is for a partial open throttle condition.

From Figure 13, the following can be observed:

  • The vehicle meets the target of 70% or higher up to 4000 rpm.
  • Near the redline (close to 6000 rpm) of the engine, the articulation index is 50%.  In practice, an Articulation Index of 70% is often difficult to meet at redline.  An Articulation Index of 60% would be considered as good.
  • The green trace (wide open throttle) is lower than the blue trace (partial open throttle).  For much of the range, the articulation index is about 8% lower when running at wide open throttle than running at partial throttle.

 

In this example, Articulation Index was used more as a indicator of the quietness of the interior, not a measure of speech interference.  However, with the emergence of voice commanded infotainment systems, speech interference is also an important consideration.

 

Note that an extremely high Articulation Index, like 100%, is not necessarily desirable.  Some blocking of speech is helpful.  For example, in an airplane, one would not necessarily want to hear a crying baby five rows back with great clarity.

 

6. Modified or Open Articulation Index

 

Open Articulation Index has a wider range than 0 to 100%. The acronym AIM (Articulation Index Modified) is sometimes used to refer to this method. 

 

The AIM method opens up the articulation window from a range of 30 dB (see Figure 4) to a fixed range of 80 dB between the limits of 20 and 100 dB. See Figure 14 for a comparison between Open Articulation Index and Articulation Index for the same time recording.

 

AI_Open_Closed.pngFigure 14: Articulation Index (blue) versus Open Articulation Index (green) for the same sound recording.

The results of this method can range from -107% to almost 160%.

 

7. Calculating Articulation Index in Simcenter Testlab

 

In Simcenter Testlab, turn on ”Signature Throughput Processing” under “Tools -> Add-ins” to calculate Articulation Index as shown in Figure 15. Signature Throughput Processing requires 36 tokens, if using token licensing.

 

AddIns.pngFigure 15: To calculate Articulation Index, turn on “Signature Throughput Processing” under “Tools -> Add-ins”.

Under Section Settings, find the “Psychoacoustic Metrics” tab and turn on “Articulation Index” as shown in Figure 16.

 

AI_Simcenter_Testlab.pngFigure 16: Turn on the checkbox next to Articulation Index in the Psychoacoustic Metrics tab.If interested in Articulation Index versus time, set the Method to Tracking on Time with a fine increment. See the Knowledge base article “Throughput Processing Tips” for more settings.

 

8. Calculating Articulation Index in Simcenter Testlab Neo

 

(view in My Videos)

 

The Simcenter Testlab Neo Process Designer can also be used to calculate Articulation Index.  The Process Designer is delivered with Simcenter Testlab Release 18 and higher, and can be run using Simcenter Testlab tokens.

 

8.1 Getting Started with Simcenter Testlab Neo Process Designer

 

In Windows, start the Process Designer by selecting “Simcenter Testlab 18 -> Testlab Neo General Processing -> Process Designer” (Figure 17).

 

Simcenter_ProcessDesigner.pngFigure 17: “Simcenter Testlab 18 -> Testlab Neo General Processing -> Process Designer”.

If the startup splash screen appears, choose the “Universal View” under Processing (Figure 18).

 

SplashScreen.pngFigure 18: Choose “Universal View” from the start splash screen.

After Simcenter Testlab Neo has started, check that the “Process Designer” and “Sound Quality Analysis” add-ins are enabled under “File -> Add-ins”.  Together they require 52 tokens. See Figure 19.

 

Neo_AddIns.pngFigure 19: Turn on “Process Designer” and “Sound Quality Analysis” under “File -> Add-ins”.

Now the processing can begin!

 

8.2 Using Simcenter Testlab Neo Process Designer

 

Simcenter Testlab Neo Process Designer has three important areas: Data Selection, Process, and Display a shown in Figure 20.

 

Neo_Functions.pngFigure 20: The Simcenter Testlab Neo Process Designer has three main functional areas: Data Selection, Processing, and Display.In the Data Selection area, navigate to the time data to be processed. Right click on it and select "Add to Input Basket".

 

In the Process area, click the down arrow in the “Add method” box and choose “Articulation Index” (Figure 21).

 

Neo_Method_Selection.pngFigure 21: Add the “Articulation Index” method to the Process area.

The entire process should have two methods: Input and Articulation Index (see left side of Figure 22). Press the Run button in the lower left to calculate the Articulation Index (see right side of Figure 22).

 

More_Neo.pngFigure 22: Left side – The complete process for Articulation Index calculation. Right side – Run button to execute process in lower left of screen.

After the processing finishes, press the “Accept” button in the lower left to save the processed Articulation Index into the project (Figure 23).

 

NeoSaveResults.pngFigure 23: Press the Accept button to store the processed Articulation Index results in the project.

Hope this helps in understanding Articulation Index and how to calculate it.

 

Questions?  Email peter.schaldenbrand@siemens.com or post a reply!

 

Related Sound Quality Links: