Category: HW2

Part 1: Research

The iPhone XR has three microphones: one in the front at the top of the touch screen, one on the rear side of the phone near the back camera, and a bottom microphone near the edge of the bottom speaker. It primarily uses the bottom microphone when recording.

The maximum sample rate is 48 kHz and bit depth is 32-bit. After researching different voice recording apps for the iPhone, I settled on Voice Record Pro. It has 9.9K reviews, and almost 5 stars, so I was hoping that that meant this is a good app to use. It allows the user to record voice memos and on-site sounds at unlimited length with configurable quality, and we can even export sounds to import from Google Drive. It can also record directly into the .WAV format, which was the requirement of this assignment. Not to mention, it’s free!

Other apps I researched include Voice Memos (of course, since it was already downloaded into our iPhones from the beginning) but I was quick to rule this out because I am unable to record in a .wav format. Another high contender was “Awesome Voice Recorder” which was said to be best for music industry professionals, and also supports .WAV format. However, I think the Voice Record Pro would have been easier to use, and I was able to set its gain control to 0 which is a requirement of the assignment.

Part 2: Data and Recording

I chose to use my MacBook Pro laptop speakers to generate the sine sweep and white noise from Audacity. I chose my independent variable, or the variable that I change, to be the placement of the iPhone XR’s bottom microphone in relation to the laptop speaker. I placed the iPhone next to the laptop speakers, which are located on either side of the keyboard, two ways: with the iPhone bottom microphone next to the laptop speaker or with it flipped around on the other side.

spectrum.txt — the sine sweep generated by Audacity’s .txt file.

For the sine sweep, as I expected, the audio recording was more stable for the iPhone microphone being right next to the speaker in contrast to it facing the opposite direction. Right at around 3000 Hz for the iPhone mic next to the speaker, the dB drops and becomes a lot bumpier than the clean sine sweep of the original audio. However, with the iPhone mic facing away in the opposite direction, the audio grab is rocky from the start and features many bumps. It start being irregular from around 300 Hz itself.

For the white noise, interestingly enough the original generation itself is supposed to be rocky in comparison to the sine sweep. Somehow, the dB in the iPhone mic seemed to be featuring a higher number than the original, but we can see the audio taper down at the end in the higher frequencies for both iPhone grabs. Again, the microphone away from the speaker is bumpier.

In terms of spectral flatness, I think that the iPhone mic, when optimized to be placed right next to a speaker or source of sound, is pretty good at having a flat response with lower frequencies up until 3000 Hz and really high frequencies. For white noise, it has more shaped response.

I will link the Google Document that has the original tables rather than screenshots here: https://docs.google.com/document/d/17ApBtmifboUHYZuteERxK-Ud8an20-QOCtMtIAt63r4/edit?usp=sharing.

The folder containing the four iPhone recorded .WAV files can be found here: https://drive.google.com/drive/folders/1CGA7ifcNeyNqp98-VYT8QKxWwDsMxCAk?usp=sharing.

What I would do differently:

I think I was really apprehensive at first to turn the laptop speaker all the way up since I was in my dorm room with suitemates who, if they could hear me playing the sine sweep, would probably be very angry with me. I tried to play it off like an ambulance, however. I would also try different variables of location, like what it would sound like in a practice room and what the audio recording’s data would look like, since my room has some ambient noise as my window faces the street. The window, however, was closed, and all my fans were off in the making of these recordings.

Citations:

https://discussions.apple.com/content/attachment/8db352cc-b4df-461b-9e14-94d0e5629571

https://apps.apple.com/us/app/voice-record-pro/id546983235

https://www.howtoisolve.com/where-is-the-microphone-on-iphone-xs-max-iphone-xs-iphone-xr-location/

Unpacking the iPhone 11 and the rationale behind 48kHz

Per Apple’s design specs and initial observations, the standard iPhone 11 has three microphones (all stereo).

Apple fails to officially report the product’s microphone’s maximum sample rates and bit depths, leading to conflicting information across discussion boards and third-party websites – however, per the most common figure and having figured out in practice, it appears that the iPhone 11 microphones are equipped with both mono and stereo input capabilities, at a maximum sampling rate of 48000Hz and bit depth of 32, progressing from the former 44.1kHz industry standard generally employed to export audio to CDs. It was simultaneously quite interesting to find that 48kHz sampling rates are becoming increasingly desirable for their compatibility with video standards.

Regaining control: Finding the right app

Finding an recording app that ensures uncompressed audio and the absence of built-in gain control was an intriguing process. 48kHz audio sampling is becoming increasingly common, allowing most standard recording apps to tick this box; although, interestingly, Apple’s own GarageBand records at 44.1 kHz.

I had a crack at a few audio recording apps: Dolby On enabled lossless exports and 48kHz/16-bit recording but seemed to suggest it had AGC in place (see below), neither Røde Reporter nor Rec seemed to suggest they could disable standard iOS gain control and Hooke Audio did not seem to support anything other than AAC/MP3 exports. As such, I opted for software developer Dayana Networks’ Voice Recorder Pro as my recording software, as this app was capable of executing each of the recording specifications of this experiment.

Putting it all into practice

I used Audacity to generate the sweep at a constant amplitude of 0.5. Using my laptop’s dual speakers, I placed my phone next to the left speaker, right speaker, at the center of the laptop, and 6ft away from the laptop (social distancing and its impact on spectral flatness?), and recorded each individual sweep using Voice Recorder Pro. After noticing negligible differences in spectral flatness between the location of the phone in regard to particular laptop speaker (i.e R/L, center), I repeated the short distance experiment outside of the suite, to observe the impact of background noise on the iPhone’s recording ability.

Here were the results:

It can be seen that the iPhone 11’s mic is able to pick up the majority of frequencies at a reasonable level; the perceived lack of flatness in parts of the FFT analysis can partially be attributed to background noise, and we can generally conclude that the spectral flatness and FFT results of the phone recording is dependent on a) the loudness/distortion of the output, b) the distance from the output (shown by our “socially distanced” recording 6ft away from the laptop and c) the room size (taking a recording in the courtyard demonstrated a noticeably quieter, “messier” and less flat recording, despite distance being similar to that of the recordings taken adjacent to speakers). Qualitatively, the lack of an oscillation in peaks demonstrates a solid spectral flatness to the iPhone 11, making its three stereo microphones of a clearly high quality and justifying android slander universally.

Sites consulted:

https://www.provideocoalition.com/48-khz-how-to-set-it-in-android-ios-macos-and-windows/#:~:text=Setting%2048%20kHz%20audio%20sampling%20in%20iOS,Airlinc%20(reviewed%20here)

https://www.provideocoalition.com/all-audio-production-distribution-should-go-48-khz-learn-why/

https://www.gizmochina.com/product/apple-iphone-11/

https://forum.juce.com/t/disable-agc-and-highpass-filter-on-ios/12812

https://www.apple.com/iphone-11/specs/

https://www.howtoisolve.com/where-is-microphone-in-iphone-11-pro-iphone-11-pro-max-and-iphone-11-exact-locations/

Apple iPhone XS – Smartphone Specs

The Apple iPhone XS smartphone released in 2018 has multiple, omnidirectional microphones for stereo to create a more reliable frequency response. The iPhone XS has stereo sound recording options. In assessing this iPhone’s features, DXOMARK describes the location of the various built-in microphones (and speakers) as situated:

• at the bottom edge of iPhone (primary microphone) with the perforations facing downwards when the iPhone is held in portrait mode,
• on the front screen’s top notch (front microphone) of the display facing outwards, and the 
• near the back camera (rear microphone) with the perforations facing downward when in portrait position.

While the phone doesn’t necessarily say which microphone is being used, functionality determines which microphone is being used when recording.

Key audio specifications (Audio Playback):

• “Stereo speakers and recording”
• “Active noise cancellation with dedicated microphone”
• “Audio formats supported: AAC‑LC, HE‑AAC, HE‑AAC v2, Protected AAC, MP3, Linear PCM, Apple Lossless, FLAC, Dolby Digital (AC‑3), Dolby Digital Plus (E‑AC‑3), Dolby Atmos, and Audible (formats 2, 3, 4, Audible Enhanced Audio, AAX, and AAX+)”
• “Wider stereo playback”
• “User‑configurable maximum volume limit”

Sample Rate and Bit Depth

In learning more about digital audio, it helps to understand that sample rate is the measure of capture and playback and the bit depth (“word length”) equals each sample’s number of bits. So, if 24 bits (three bytes) means that there are 24 binary digits per word, a higher bit depth results in a more refined measurement. Ultimately, more data allows for better recreation of sound. Conversely, all of the data cannot be gathered and reproduced and the information degraded if the bit depth is too low. According to PreSonus.com, “for perspective, each sample recorded at 16-bit resolution can contain any one of 65,536 unique values (2^16). With 24-bit resolution, you get 16,777,216 unique values (2^24)—a huge difference!” in accurately recreating sound.

 

Approaching and Approving Apps

Before you can record anything, you need to decide what app to record in.

(1.) The first app I tried out was Apple’s “Voice Memos” recording app, a default iOS app installed on all iPhones automatically. I’ve used this app a lot in the past—mostly since it came with the phone. In general, the iOS Voice Memos app is quick and easy to record sound in; however, for the purposes of this experiment, the Voice Memos app isn’t the greatest. It doesn’t display the qualities of its microphones used to record, nor does it display most of the properties of any given recording.

(2.) I then went on to check out (2.) TwistedWave’s “TW Recorder”, an app that classmate Michael Lee recommended. I downloaded a free version of the iOS TwistedWave Recorder app, and it worked great! It allows the user to see a lot of the properties of a recording, adjust your recording settings and preferences, save in multiple different formats, and set a new recording’s properties: like its sample rate (either 8, 11, 16, 22, 32, 44, 48, 64, 88, or 96 kHz) or whether it will be recorded in Mono or Stereo.

(Apple.com – TwistedWave Recorder)

While Apple does not officially publish their phones’ maximum sample rate and bit depth, I used the TwistedWave app to assess the quality of input/output sound for this experiment. I used the app’s settings to determine while sampling rates were unsupported (and, thus, figure out the maximum). 96 kHz, 88 kHz, and 64 kHz were unsupported on the app, so I used this to find out that 48 kHz is the maximum sample rate of the iPhone XS (lining up with what external websites tended to say, as well).

However, the correct maximum bit rate was a bit harder to find. From external websites, it was unclear whether it was 16-bit, 24-bit, or 32-bit. While the TwistedWave app says that it’s able to store recordings in 32-bit, I decided to stick with recording with 16-bit, which I knew the phone could record.

According to the MacWorld review, a significant sound quality feature of the Apple iPhone XS includes “record[ing] sound in stereo, and audio played from the XS [has] better clarity and volume”. 

 

Generation of Sine Sweep and White Noise

I used Audacity to generate two audio clips: a 15-second sine sweep rising from 100 to 18,000 Hz, and a 15-second clip of white noise.

I then used my phone to record these sounds. All of the sounds were recorded in my bedroom (in consideration to my suitemates studying in the suite’s common room), but with my phone placed in different locations in relation to the computer and its speakers.

I used Audacity’s “Plot Spectrum” functionality on the original sounds and my phone’s different recordings (as in the table shown below). The spectrum plots show Hz frequencies on the horizontal axis and dB of that frequency on the vertical axis.

FREQUENCY RESPONSE GRAPHS (Sine Sweep):

Place for Recording	Phone Placement	Type of Test	Plotted Graph	Results of Test
Audacity	original sound from Audacity	Sine Sweep		Flat
bedroom	1 inch above left speaker	Sine Sweep		Shaped
bedroom	1 inch above right speaker	Sine Sweep		Shaped
bedroom	held 1 inch above speakers; centered between them (5 inches on either side)	Sine Sweep		Shaped
bedroom	12 inches away from speakers	Sine Sweep		Shaped
bedroom	4 feet away from speakers	Sine Sweep		Shaped

 

FREQUENCY RESPONSE GRAPHS (White Noise):

Place for Recording	Phone Placement	Type of Test	Plotted Graph	Results of Test
Audacity	original sound from Audacity	White Noise		Flat(!)
bedroom	1 inch above right speaker	White Noise		Shaped
bedroom	1 inch above right speaker	White Noise		Shaped
bedroom	held 1 inch above speakers; centered between them (5 inches on either side)	White Noise		Shaped
bedroom	12 inches away from speakers	White Noise		Shaped
bedroom	4 feet away from speakers	White Noise		Shaped

Recording Using iPhone XS Microphone: Analysis and Understanding

Using the plot of the original Audacity sounds as the “ideal” benchmark for audio capability, the white noise and sine sweep graphs reflected a change in audio quality from the Audacity baseline model. Amongst the sine sweep graphs and white noise graphs, generally-similar shaped patterns emerged when the sound source remained one inch above either speaker (either left or right). However, once the phone was moved more than a few inches away from the speakers, the graphs reflected less spectral flatness.

Again, with the Audacity plot as the desired baseline graph for spectral flatness, the sine sweep and white noise graphs indicated a degraded shift in audio quality based on the graphs’ increased area and filled-in peaks and valleys. Unlike the plots of the phone’s recordings—both when the microphone was close and far away from the output source—the Audacity plot maintained consistent peak heights throughout the graph translating to a more consistent frequency response. In the recording graphs, the number of decibels fluctuated.

In a majority of the recordings, the frequencies above 2000 Hz tend to have slightly higher dB than the frequencies below 1000 Hz. A majority of the recordings have dips in dB at some point in the range between ~1000 Hz and ~1500 Hz. For example, both of the two recordings taken about 12 inches away from the speakers dip in the range between 1000 Hz and 1300 Hz.

Both of the two recordings taken about 4 feet away from the speakers dip fairly significantly between 9200 Hz and 11,200 Hz.
In the recordings centered between the two speakers (5 inches away from either speaker, and held 1 inch above them), the dB seems to slightly dip starting at 300 Hz, increase again starting at 400 Hz, and peak at 500 Hz.
The spectrum plot of the original white noise audio at first looks quite erratic, uneven, and shaped, but the graph is actually zoomed in to a small range of dB, between -28.2 and -29.5. If it was shown at the same scale as the others, it would appear quite flat and consistent.

The original sine sweep audio only went from 100 Hz to 18,000 Hz, so it makes sense that the five recordings of it also cut out at around 18,000 Hz.
However, the original white noise audio went up to 23,500 Hz, so we can use this to see where the iPhone XS’s limits are with recording high frequencies above 18,000 Hz.

• Apple designed and built the iPhone XS with omnidirectional microphones to create a reliable response frequency, which it does. To achieve an excellent frequency response, Signal Essence indicates that “a flat frequency response from 20Hz to 20kHz” is ideal and “an iPhone [is] pretty flat from 100Hz to 10kHz [which is] good enough” and meets the design goal of a reliable response frequency.
  
• All of the recordings begin to drop out in volume around 20,000 Hz (as expected), and seem to drop out completely shortly afterward, at slightly different Hz frequencies. (For the purposes of this analysis, I’m counting the point of “fully dropping out” as falling below –100 Hz.)
• In the recording centered between the two speakers (5 inches away from either speaker, and held 1 inch above them), it appears to have fully dropped out at around 20,150 Hz.
• In the recording 12 inches away from the speakers and the recording 4 feet away, it appears to have fully dropped out at around 20,200 Hz.
• In the two recordings 1 inch directly above either speaker (left or right), it appears to have fully dropped out at around 20,350 Hz.

Consequently, in assessing the spectral flatness of the iPhone XS’s recordings, I found that this device’s microphone does have a generally reliable frequency response— able to pick up a range of frequencies as intended, designed, and built. However, as expected, both the environment and relative distance from the sound source both impacted the spectra. As reflected in the graphs and observations,  the closer the sound output source to the microphone, the more spectral flatness (generally translating to better quality and clarity); conversely, further distance from the sound source created more variation and inconsistency. Overall, my assessment of the iPhone XS’s microphone’s audio capability performed with a generally reliable frequency response.

Since Apple doesn’t tend to publish some details of its devices’ specifications, research is required to unearth the phone’s capabilities — but, once tested, it lives up to expectations.

Or, essentially, assessing “excessively extensive testing” successfully expresses XS’s expected impressive technical specs. Success!

 

Recordings: https://drive.google.com/file/d/13sPvDvk7MFhi3CBAwfRmDudeCzBlgruy/view

Sources

Apple.com – Discussion thread – Where are the microphones on an iPhone XS/Max?
https://discussions.apple.com/thread/8555709

Apple.com – iPhone XS – Technical Specifications
https://support.apple.com/kb/SP779?locale=en_US

Apple.com – TwistedWave Recorder
https://apps.apple.com/us/app/twistedwave-recorder/id690359266

The CellGuide.com – How to fix Apple iPhone XS microphone that is not working
https://thecellguide.com/how-to-fix-apple-iphone-xs-microphone-that-is-not-working-5892

DXOMark.com – Apple iPhone XS Max Audio review
https://www.dxomark.com/apple-iphone-xs-max-audio-review/

gmsarena.com – Apple iPhone XS pictures
https://www.gsmarena.com/apple_iphone_xs-pictures-9318.php

iKream.com – How to fix Apple iPhone XS microphone problems, microphone not working [Troubleshooting Guide]
https://www.ikream.com/iphone-xs-microphone-problems-microphone-not-working-30022

MacWorld.com – iPhone XS and iPhone XS Max review: It’s time for older iPhone owners to jump on the X bandwagon
https://www.macworld.com/article/3309412/iphone-xs-and-iphone-xs-max-review.html

PreSonus.com – Digital Audio Basics: Sample Rate and Bit Depth
https://www.presonus.com/learn/technical-articles/sample-rate-and-bit-depth

PhoneArena.com – Apple iPhone XS Description
https://www.phonearena.com/phones/Apple-iPhone-XS_id10766

SignalEssence.com – Can you use an iPhones’ internal microphone for acoustic testing and accurate recordings
https://signalessence.com/can-you-use-an-iphones-internal-microphone-for-acoustic-testing-and-accurate-recordings/

StackExchange.com – What is the maximum sample rate/bit depth of the iPhone for recording?
https://sound.stackexchange.com/questions/15226/what-is-the-maximum-sample-rate-bit-depth-of-the-iphone-for-recording

Initial Research

For this project, I was equipped with my trusty iPhone 8. Although Apple doesn’t publish any specs on its microphones, plenty can be determined with a little bit of digging. Several articles point to the iPhones XS and XR as the first Apple phones with stereo microphones, so my iPhone 8 will be restricted to mono (PhoneArena). Another article measures the frequency response and dynamic range compression of the iPhones 6s, 7, and X. While my phone wasn’t included in the study, the measurements give a pretty good picture of what we should be dealing with, as the iPhone 8 was released between iPhones 7 and X.

Frequency response

When it comes to frequency response, all three phones sport mostly flat profiles, save a roll off around 50-60Hz at the low end, and some funky behavior around 10kHz. All should record perfectly fine in the 100Hz to several-kHz range, but we may expect to see some distortion in the upper half of the (linear) spectrum (SignalEssence).

Dynamic Range Compression

To make sure the iPhones wouldn’t compress the dynamic range of loud samples, this study recorded samples at 90dBSPLA, 96dBSPLA, and 99dBSPLA (dB SPL A-weighted). All three phones elicited similar results, so the three lines on the graph show the loudness of the recorded audio for each sample. While the article notes that the iPhones began clipping above 99dBSPLA, the 96dBSPLA and 99dBSPLA samples were recorded with accurate dynamics relative to the 90dBSPLA reference (0dB input gain). This bodes well for our iPhone 8 (SignalEssence).

Recording Software

It wasn’t until I looked into the recording software itself that I could find out the specific bit-depth and sample rate my phone would be capable of. After some searching on google, I landed on the app Auphonic (Available on the iOS App Store). This free app presents a simple interface with all the settings we need to control, so choosing it was a no-brainer. The main interface of the app looks like this:

After looking through these settings, I found that my iPhone was capable of 24bit/48kHz on 3 (!) different microphones (front, bottom, back). To ensure Auto Gain Control (AGC) would remain off, I selected -12dB gain (the closest option to 0dB). With my research and setup complete, it was finally time to start recording!

Generating sounds

Actually, not so fast. Before I could record anything, I had to produce the sounds I was going to play: 15 seconds each of a sine sweep (100Hz-18kHz) and white noise. For this task, I turned to the application Ocenaudio, available on macOS, Windows, and Linux (website). The app has an intuitive interface that made everything super easy!

After opening Ocenaudio, I selected Generation>Tones to generate the sine sweep. After selecting options for audio quality (I selected 24bit/48kHz mono), I found the tone generation menu. With the settings pictured above, I pressed “OK”, then exported the file to full quality WAV (for playback) and 128kbit/s CBR MP3 (for web upload), and completed an FFT analysis on the clip (to be used later). The process was much the same to generate the white noise.

Recording

Now equipped with my audio files, I was ready to record them to see how my iPhone would perform. To spice it up a little, I decided to compare all three microphones on the device, each at a 2ft distance from my laptop speakers (MacBook Pro 16″, 50% volume).

Results

After recording all 6 clips ( [front, back, bottom]x[sine sweep, white noise] ), I uploaded them to ocenaudio on my computer, cut the leading and trailing silence, exported to MP3 (for web upload), and performed an FFT analysis. All audio clips and graphs are shown below.

Sine Sweep

Sine sweep (original):

Sine sweep (front mic):

Sin sweep (back mic):

Sine sweep (bottom mic):

Analysis: The intensity of the original sine sweep is flat over the 100Hz—18kHz range, until sloping down to lower intensities at 18kHz and above. I’m not exactly sure where these higher frequencies came from, but they seemed to be be picked up only by the back mic, while the bottom and front mics display a full cutoff at 18kHz. In regards to the shape of the recordings, the front mic seemed to capture a flat spectrum, while the intensity recorded by the back and bottom mics decreased asymptotically as frequency increased. Perhaps these decreasing profiles have something to do with lower frequencies being perceived as louder or echoing more in the room.

White Noise

White noise (original):

White noise (front mic):

White noise (back mic):

White noise (bottom mic):

Analysis: While the initial recording is flat across the full frequency spectrum (low Hz to >25kHz), both the back and bottom microphones cut of around 20kHz, indicating either that they aren’t responsive to higher frequencies, or something about the room / distance from the speaker muffled out the high end. Meanwhile, the front mic recorded more lows (<1kHz)—though this likely could have been simply due to movement of the phone during recording.

Notes

For the life of me, I couldn’t figure out how to get the images to display next to each other. After the “align-left” jumbled with more than three images, I even tried injecting some CSS into all the <img> elements (to no avail). Does anyone know how to do this?

Works Cited

PhoneArena https://www.phonearena.com/news/Apple-iPhone-XS-MAx-records-stereo-sound-with-four-mics_id108769 

SignalEssence https://signalessence.com/can-you-use-an-iphones-internal-microphone-for-acoustic-testing-and-accurate-recordings/

Experimenting with Recording on the iPhone 7

 

Suffice to say, I haven’t gotten a new phone in almost 4 years, and the phone that I do have has been battered and bruised in all sorts of ways. So I was a little skeptical at first about its recording capabilities. But after some digging through online forums and archives (as Apple, annoyingly, removes specs for any products that they no longer sell), I was pleasantly surprised by the microphone specs of the iPhone 7.

 

The iPhone 7 has 4 microphones in total: two on the bottom on either side of the Lightning charger port, one on the back next to the rear camera, and one inside the built-in earpiece. The 4 mics all work together for purposes of noise cancellation and beamforming (locating the source of a sound), and all can be used to record (with the iOS automatically selecting the most appropriate microphone for recording, such as the rear mic for shooting video). The microphones can all record in stereo, and with default Apple apps, are limited to recording at 16-bit with a maximum sample rate of 48 kHz. 

 

Researching Recording Apps

 

Finding a suitable recording app was definitely challenging, as the only app I’ve ever used to record is Voice Memos. Apparently, Voice Memos settings can be changed to produce lossless audio, but not at the maximum sample rate, not in a WAV/AIF format, and with no AGC control. Thus, I turned to Google, eventually finding two possible apps on a birdwatching forum, out of all places. I definitely trusted these apps, as the quality and precision of audio required to record bird sounds in the wild are far higher than recording sine waves in my bedroom. 

The two apps recommended throughout the forums were RØDE Rec LE and Voice Record Pro, which both can produce lossless audio as WAV files, at 48k/16-bit, and have options to disable auto gain control. After perusing the app store, I found that RØDE Rec LE had absolutely abysmal reviews, mostly because of its disgustingly archaic interface and lack of flexibility in export options, so I turned quickly to Voice Record Pro. This turned out to be the right choice, as Voice Record Pro was clean, easy to use, and extremely functional in customizing the presets for this project (pictured below).

Data Collection and Observations

 

In Audacity, I first generated 15 seconds of white noise at a constant amplitude of 0.8, then a 15-second sine sweep/chirp at a constant amplitude of 0.5, ranging from 100 Hz to 18 kHz. Then, using Voice Record Pro and my pre-configured settings, I recorded both sounds at maximum volume, holding the microphones on the bottom of my phone towards my computer speakers, at about 6 inches away.

I recorded in my dorm room, which was the only feasible location I had, as recording sine waves and white noise in the library probably would’ve been a little disruptive. With my door closed, the background noise generally was at a minimum, except for when my new suitemate decided to move into his room right in the middle of recording. The playback of the audio was also decent, even through my beaten up, five-year-old Macbook Air speakers.

 

I also varied two other factors to see their effects on the resulting spectra. First, I varied the volume of the sound that I played from my computer. Using the arbitrary bar that shows up when you adjust the volume on a Macbook, I had originally played the sound at a maximum 16 bars. Now, I played it at 5 bars, then 10 bars. These were the resulting spectra:

Then, I varied the direction that I faced the microphone at the bottom of my phone. Originally, I had faced it directly towards the computer. Now, I faced it away from the computer (towards me), then downwards at the computer keyboard.

These were the resulting spectra:

 

My final observations from the data were pretty straightforward and expected. The lower the volume of the playback was, the more erratic and uneven the curves were. The greater the volume, the more even, as sound was most likely more consistently reaching the microphone. The direction that I faced the microphone didn’t make as much of a difference as I thought it would, but in general, facing the microphone straight at the speakers intuitively produced a louder and more even curve. In general, probably as expected, I would say the most consistent, “best” recording comes when the microphone is directly facing the speaker, and the playback volume at the maximum. This, however, is a fairly presumptuous statement that probably would require more extensive testing and more variables to completely prove. And in the end, even though the iPhone 7 clearly was not close to replicating the original generated sounds, when recorded in the right manner and in the right environment, it certainly can achieve decent standards of recording quality.

For the original, generated audio files and all the recorded audio files, along with the exported spectra txt files for each recording, please see this Google Drive link: https://drive.google.com/drive/folders/1PDsPzO_fW1rauO1aDrZet5JyQPlVv4Zt?usp=sharing

 

Sources

https://apple.stackexchange.com/questions/251930/wheres-iphone-7-microphone-located#:~:text=The%20iPhone%207%20has%20four,the%20speaker%20grille%20as%20well.

https://apple.stackexchange.com/questions/276905/what-audio-quality-is-available-on-newer-iphones#:~:text=There%20appears%20to%20be%20no,16%2Dbit%20at%2048%20kHz.

http://www.tomasdabas.eu/tutorials/record-quality-voice-over-using-smartphone/

http://www.rode.com/software/roderecle

https://www.bejbej.ca/app/voicerecordpro

https://support.ebird.org/en/support/solutions/articles/48001064305-smartphone-recording-tips

https://www.macaulaylibrary.org/resources/setting-up-recording-apps/setting-up-recording-apps-for-ios-devices/

Research

Unfortunately, Apple does not release information or specs of their devices on their website. There are third parties websites with what seem like credible information, but even then the Iphone 6s is an outdated model that can be perceived as obsolete face to the new Iphone models. What I can say, is that the Iphone 6s has 3 microphones: one on the back of the phone by the camera lens, one on the front of the phone by the camera lens, as well as two on the bottom of the phone (on the left). The phone, according to Apple, should have somewhat of a flat frequency response from 10Hz to 10kHz, for recording speech. I was unable to find any other pertinent information.

Record

The app used to record the sine sweep was: Auphonic Recorder

The app allows the user to:

• Turn the gain off or on
• Choose which microphone utilize
• It allows a choice of format (AAC audio or PCM/ WAV audio)
• Choice of precision of recording (16 or 24 bits)
• Allows a sample rate of up to 48kHz

For this Recording the specific setting chosen were:

• Gain: off
• Format: wav
• Sample rate: 48 kHz

The program I used to analyze the sweep was Oceanaudio. I analyzed a 15 second recording on FFT with a linear scale.

At first, I used my BOSE speaker to record the sine sweep with my phone about a foot from the microphone (The volume of the speaker was actually pretty loud). After using Oceanaudio to analyze the recording I got this:

[could not manage to uplaod graph]

According to the FFT analysis, we can see that the frequency response of built-in Iphone 6s microphone (bottom one) seems to be not very flat at all after passing the 1kHz mark. Before then, the frequency response is quite flat. However, after passing the 10k Hz range there are significant dips in the frequencies (for example at 6kHz, 8.5 kHz or 11 kHz, etc…). I wanted to try recording again but at a closer distance of the phone to the source to see if the dips were caused by outside faint sounds.

I did another recording, this time putting my phone between by Sennheiser headphones so that the Iphone 6s microphone could be as close as possible to the source. This time the FFT analysis provided this:

[could not manage to uplaod graph]

Here we see by actually moving the phone closer the frequency response of the sweep is flatter from 100 Hz to 9kHz but dips down significantly after 10kHz.

Report

The Iphone 6s is an older Iphone model which does not reproduce sound as well as the newer models and as other phones, when recording at a distance it is inaccurate. It does however reproduce sound very well when the sound is very close and in the 10 Hz to 10 kHz range. This shows how the Iphone 6s is not a phone meant to record music or singing, it is strictly a phone created to talk and used to call and clearly hear people on phone calls (speech is very much in the 10 Hz to10kHz rang. This all makes sense for an older model Iphone giving up microphone recording capabilities at various frequencies in order to capitalize on the frequencies used to communicate through phone calls.

The iPhone SE (2020)

My iPhone SE has two built-in microphones (one at the top, and one at the bottom) and stereo playback. The bit depth is 24 bits and it has 48 kHz playback. Interestingly, “left and right channels are inverted when playing back music in landscape mode” (dxomark.com).

Finding an App

I found “Voice Record Pro” on https://www.popsci.com/record-better-smartphone-audio/, an app that allows both manual gain control and an option to enable iOS’s automatic gain control. Before starting a recording, it allows you to set various settings such as sample rate, bit depth (max. 32 bits), channels (ex. stereo), encode quality, and record format (the last of which include .mp3, .mp4, and .wav, among others). In addition to the settings shown in the first picture (below, left), there is an additional screen of settings so that the user can customize their experience (below, right).

      

I was also considering Spire, which had good reviews on the App Store. In addition, an article described the gain input as adjustable, but I wasn’t able to find the setting on the app. Overall, Spire has a sleeker interface but fewer options for audio settings, so I used Voice Record Pro.

Recording + Analysis

I recorded in two locations: my suite’s common room (with the windows and doors closed to minimize background noise; has some furniture but overall pretty bare) and my closet (very small; has a good amount of clothes hanging). In the common room, I had three distances: near (phone right next to the computer, with the bottom of the phone closest to the computer speaker); medium (phone and computer about 2/3 of the width of the room away); and far (maximum distance away in the common room – computer in one corner on the floor, and phone in the opposite corner on a high shelf). In the closet, I had two distances: near (defined same as the common room) and medium (defined as the common room’s “far” – in two opposite corners; I called it medium because the distance between the phone and computer was significantly shorter than the “far” I used in the common room).

The record settings I used are the ones shown in the picture (above, left) with the exception of bit depth, which I changed to 24 bits. There was definitely some background noise, especially for the common room recordings — at least twice a car drove by, making a noticeable sound.

Recordings & Graphs

Recordings can be found here; the Google Doc with the tables + graphs is here.

Sine Sweep

Analysis

• The “common room, near” looked the most similar to the original graph, though the “common room, near” graph tapered off at the end (whereas the original graph stayed around the same level)
• I was surprised by how different the “common room, near” looked in comparison to the “common room, medium” and “common room, far”; in addition, I would have expected the “common room, near” and the “closet, near” to look relatively similar (as they do for white noise – graphs below).
• All 5 graphs seemed to reach a relative minimum (in dB) around 700-800 Hz. Before this point, the graphs were relatively smooth; after this point, the graphs became “bumpier” (more variation in the same time span). The original graph seemed to reach a relative minimum slightly later, around 800-900 Hz.
• The peaks for the “common room” recordings varied between all five recordings. The two closet recordings’ dB levels both had relative maxima around 400 Hz, but the “closet, near” recording actually became louder at the end — I would guess that this was from background noise.
• Overall, there is relative spectral flatness (especially compared to the white noise graphs)

White Noise

Analysis

• For the white noise, the overall shape of the graphs of the three recordings from the common room seemed more similar – though the beginnings were still slightly different
• The differences between the original and these graphs seem greater than the differences between the the original and my graphs for the sine sweep (especially comparing the original graph and the “common room, near” graph)
• The two “near” recordings (one in the common room and one in the closet) had the most spectral flatness; the other three varied significantly more
• These white noise graphs seem to have more fluctuation than the sine sweep graphs

Overall

• The “near” recordings tended to have the strongest recording quality and be the most similar to the original graph; in addition, they had the most spectral flatness
• However, overall my phone’s recordings varied significantly from the generated recording, suggesting that factors such as background noise influenced the recording quality

Sources

De Hillerin, Marie Georgescu. “Apple IPhone SE (2020) Audio Review.” DXOMARK, DXOMARK, 29 June 2020, www.dxomark.com/apple-iphone-se-2020-audio-review/.

Smartphone Specs

The Samsung Galaxy A50 is equipped with two microphones – a front microphone and a back microphone. Both of these microphones can be used for recording (depending on the software in use). If Samsung’s built-in software called ‘Voice Recorder’ is used, only one of the microphones is utilized. The other microphone is meant to boost noise cancellation. However, softwares like RecForge II – Audio Recorder give you the option to choose between the two microphones. The Galaxy smartphone also offers stereo input options. Stereo input options can be accessed via RecForge’s ‘Audio Record’ settings. The options available are Mono and Stereo(Mono x 2). The maximum sample rate for both microphones is 48kHz. Due to insufficient information on smartphone specification websites, the audio samples were loaded on QuickTime player and were inspected using ‘Movie Inspector’. Movie Inspector showed the format of the audio samples as ‘16-bit little-endian signed integer’ from which it can be deduced that the bit depth is 16.  

 

Audio Recording Applications 

In the pursuit for an Android recording app which records uncompressed audio in .wav and disables AGC, one may stumble upon the following: Tape Machine, Smart Recorder, and RecForge II. 

While Tape Machine has a heavy fan-base on developer blogs and forums, it isn’t available from download anymore. It was hailed for having the AGC disablement feature. 

Smart Recorder is best at what it’s meant for – recording audio. Though it allows the user to disable AGC and records in .wav, the sampling rate maxes out at 44.1Hz. Apart from this it also doesn’t offer many basic audio editing features. 

In comes RecForge II – a well designed, user friendly software, allowing users to disable AGC and set a sampling rate to the maximum limit of their hardware. The application supports nearly all audio formats and offers a range of various editing tools including (but not limited to) Acoustic Echo Canceler and Skip silence. Going forward, RecForge II has been used alongside the frontal microphone to record all audio files. 

Generation of Sine Sweep and White Noise

The software used was Audacity. First, In Audacity’s Generate menu a Chirp was produced in the shape of a sine wave for 15 seconds. Next, using the same Generate menu, the Noise plug-in was used to generate 15 seconds of white noise. 

Recording Using Phone Microphone

For each 15-second sound, 3 recordings were made using the phone (six recordings in total). For the first recording the phone was placed on the laptop’s keyboard. In the second recording the phone was kept approximately 8 cm away from the laptop’s keyboard. Finally, in the third recording the laptop was placed inside a cupboard (a restricted space) and the phone was placed on the laptop’s keyboard. Gain used for all recordings was +9.0dB. 

Analysis and Understanding

Using ‘plot spectrum’ function in audacity, the following graphs were plotted for sine sweep.

All of the sine sweep graphs have one thing in common – they consume a healthy portion of the plot. The first plot is consistent and flat in terms of the relative responsiveness (expressed in dB). However, when a phone is used, the plots change and a deformed shape emerges. A change in the range of dB clearly shows that there has been a change in sound quality. When the phone is closer to the keyboard, a greater number of peaks are observed at higher frequencies. In a way we could say that being close to the keyboard ‘boosts’ the responsiveness of the microphone (same goes for enclosed locations such as cupboards). If you listen to the recordings, the recording done on the keyboard is much louder and clearer than the recording done 8cm away from the keyboard. However, the recording done in the cupboard is a louder, amplified version of the recording done on the keyboard. This can clearly be seen in the graph as at higher frequencies the 4th graph has more peaks than the 2nd graph. 

Using ‘plot spectrum’ function in audacity, the following graphs were plotted for white noise.

Compared to the sine sweep plots, the white noise graphs consume less area on the plot. The first white noise plot has various peaks however they are much lower than the flat crest observed for the initial sine sweep. By the introduction of taller peaks in the phone recorded graphs we can observe that the phone recording has severely impacted the sound quality of the white noise. Such a drastic change in peak height was not observed in the plots for sine sweep. For white noise, the change in loudness is the same as sine sweep in the sense that the more enclosed the environment and the closer the microphone to the sound source, the louder the recording. This can be observed by how the number of peaks at higher frequencies tend to get more in number and taller as the microphone is placed in a cupboard or is moved closer to the sound source. 

To draw a conclusion we can say that the loudness of the recording is inversely proportional to the size of the room and the distance of the microphone from the sound source. However at the same time the loudness is directly proportional to the number and height of the peaks at higher frequencies.  

As far as the spectral flatness of the microphone is considered overall, we can say that the microphone produced ‘shaped’ curves. This suggests that the microphone is more sensitive to certain changes in frequencies than others. This can be seen by the fact that lower frequency areas tend to have fewer peaks, whereas higher frequencies portray more peaks, clearly showing that the microphone is more responsive to higher frequencies (typically above 3000Hz, as seen on all graphs). 

Recordings: https://drive.google.com/file/d/19jfF4ezgNVOPLkGrbufHGzcD6oLoKwju/view?usp=sharing

Sources
https://forum.xda-developers.com/showthread.php?t=1042051

https://www.shure.com/en-US/performance-production/louder/mic-basics-frequency-response

https://filmora.wondershare.com/audio-editor/best-voice-recording-apps-android.html

https://support.ebird.org/en/support/solutions/articles/48001064305-smartphone-recording-tips

https://www.neumann.com/homestudio/en/how-does-frequency-response-relate-to-sound

https://www.gsmarena.com/samsung_galaxy_a50-9554.php

https://www.thepodcasthost.com/recording-skills/best-audio-recording-apps-for-android/

 

 

 

The iPhone 11 that I used for this experiment has three stereo microphones — one for phone calls (bottom), one for videos (next to the camera), and one for Siri (next to the ear speaker). Overall, reviews for the recording capabilities of the iPhone 11 are decent. The phone seems to have a pretty flat frequency response curve, excluding very high ranges. Like most devices, midrange resonances can mess up the recording quite easily, however. To my knowledge, these microphones can record at a maximum sample rate 48kz at 32-bit depth and have both mono and stereo recording abilities. For a more in-depth review of the iPhone 11 and its cousins, follow this link: https://www.dxomark.com/apple-iphone-11-audio-review/#:~:text=Like%20the%2011%20Pro%20Max,of%20bass%20impaired%20the%20sound.

The Application:

My search for a “lossless” recording application that also allowed the disabling of auto gain balance was not easy. Previously, I had regularly used Voice Memos for recordings, because I wasn’t too concerned with the audio quality at the time. Voice Memos automatically compresses audio recordings for better storage, but by navigating to the audio quality settings in General settings, one can switch on “lossless.” That being said, Memos records in mono, and there is no convenient way to regulate the Auto Gain Control. Other applications, such as Dolby On, allow lossless, uncompressed, 16-bit recordings, but seem to have various gain balancing and tone balancing technologies. I finally chose Auphonic, an application that allows users to select between the three iPhone microphones, as well as select the sample rate, bit depth, and input gain. By default, Auphonic rids all IOS preprocessing.

Recording:

I did all my recordings with the bottom microphone at a sample rate of 48khz with 24-bit (the max) precision. I used my computer speakers to project both the sine and white noise files.

My experiment included six recordings total — three for sine sweep and three for white noise. Throughout my six trials, I experimented with both room size and microphone distance from the output source. For the first two recordings of each sound, I recorded in my room at two distances — one foot away from the source and four feet away from the source. The final trial for each sound is recorded in my common room, when the microphone is 15 ft away.

FREQUENCY RESPONSE GRAPHS:

Sine Sweep

WHITE NOISE

Results:

Although there are many differences between all three White Noise recordings, some patterns established themselves. For all three recordings, The microphone didn’t really pick up any frequencies between the 200-300Hz range. The extremity of this trough seems to have been exaggerated as I moved the microphone further away from the output source.  For all three recordings, there is also a trough at the 3000-6000Hz range and a peak at 7000Hz. One can also see that my iPhone won’t capture frequencies well past 20000Hz. The good news is that the dB difference between the loudest and softest frequencies remains about the same.

There are also some recognizable patterns among the three sine wave recordings. Like the white noise recordings, there is a loss of frequencies in the 200-300hz range. There is another trough in the 700ishHz range for all three and the peaks all occur at the same frequency. On top of that, the dB difference between the loudest and softest frequencies remained similar at different distances.

From my observations, I am convinced that my iPhone 11 has a fairly flat frequency response. I am somewhat convinced that although my microphone may exaggerate the lack of 200-300hz frequencies, the problem originates from my poor computer speakers, rather than from my iPhone mic. It is also generally the case that my frequency response curves were more flat the closer to the source I was.

 

Research on Tech Specifications

My smartphone model is an iPhone 11. On researching the iPhone 11’s microphone specifications, I had to go onto several forums to gather pieces of information in addition to the official Apple site. Diagrams on the Apple site indicate three locations for the microphones. One is at the top-front of the smartphone, used for recording sound with video in real time. The other is at the earpiece, which is used for noise cancellation when making phone calls. The last two, (most important for the research) are at the bottom of the phone next to the charging port.

Other specifications found on the Apple site include the following:

• Audio formats supported: AAC‑LC, HE‑AAC, HE‑AAC v2, Protected AAC, MP3, Linear PCM, Apple Lossless, FLAC, Dolby Digital (AC‑3), Dolby Digital Plus (E‑AC‑3), Dolby Atmos, and Audible (formats 2, 3, 4, Audible Enhanced Audio, AAX, and AAX+)
• Spatial audio playback
• User‑configurable maximum volume limit

The iPhone’s standard Voice Memos is the default app for audio recording. It does not use wav files, since high quality audio is deemed unnecessary for ordinary users. Through arduous searches on the web and the App Store, I finally found an app called TwistWave Recorder, which allows for recording in wav. This app also allows the option to disable iOS control gain, which I did. 

Information on sampling rate is not available on the Apple site. When making a new file with TW Recorder, I had the option to choose my sampling rate. It was through trial and error that I found that my phone’s sampling rate is 48 kHz. I also learned this through several forums, and this app confirms it. Anything above 48 kHz and the app would alert me with a warning. 

Since about 20 khz is the max for human hearing, anything drastically higher than 40 for a sampling rate obviously would not make sense since our ears would not be able to detect it. For this reason, many iPhones have 48 kHz as the max sampling rate.

Information about bit depth is not available either. TW recorder has an option for 32-bit temporary files, with better quality but larger size (twice as much). According to the app, the default is 16-bit. I am still not certain that the bit-depth is 16. Some indicate that newer iPhones has 24-bit depth capability, but the 32-bit depth FFT graphs showed little effect in my results, so I decided to stick with 16-bit depth.

The phone also has both stereo and mono options, but I kept the option set as mono. The sweep and noise from Audacity were also generated at mono setting.

Recording Procedure

I had two variables when recording the wave and white noise with my phone. The first was the mic distance from the phone. I placed the bottom mics on the laptop speaker for the first recording, then I flipped the phone upside down for a greater distance (the top mic is not used for recording), then held the phone on top of the open laptop screen (perpendicular to the webcam) for the maximum distance. The second variable was the bit depths. Since I activated the 32-bit temporary file option, I was able to export my audio files in 32, 16 and 8 bits. For sine wave recordings I had 9 files and for white noise recordings I also had 9 files.

Constants for the recordings were the speaker used (laptop), the sampling rate (kept at 48 kHz), and the mono option as said before. Audacity was used to generate the sine sweep, white noise, and analyze my recordings.

Results and Analysis

Figure 3. Input sine sweep (top) compared with 16-bit (center) and 8-bit (bottom) recorded at the laptop speaker

The spectral flatness for these sine sweep graphs are not even close compared to the input graph in Figure 3. While this drastic difference could indicate that the microphones of the iphone may not be the best, I speculate that this problem is a result of the imperfection in the laptop speakers and environmental interference. I was in my dorm bedroom for my recordings. The room is far from being a good recording studio. There are windows, cracks in the doorframe leading to the common room, all those things come into play. 

Figure 4. White noise input from the computer (top) recorded next to the speaker at 16-bit (center) and 8-bit (bottom)

The spectral shape for these white noise graphs are relatively the same in the lower frequency range (Figure 4). The sound was much more “crunched” in playback for 8-bit. However, in both sine sweep and white noise cases, bit depth does not matter too much. 

Figure 5. Sine sweep recorded at 16-bit at distances of one phone length away (top) and one arm’s length (bottom) away from the speaker

Figure 6. White noise recorded at 16-bit at distances of one phone length away (top) and one arm’s length (bottom) away from the speaker

Distance seems to be more important than bit depth in determining recording quality.The further the distance from the speaker, the more likely the recording would be interfered by environmental noise. In both Figures 5 and 6, there is more distortion at lower frequencies. 

In summary, at lower frequencies from 50 to 1000 Hz, mainly when the mic is at the speaker, the FFT plots do not show severe fluctuations, indicating reasonable spectral flatness. However, all the plots fluctuate aggressively compared to the computer input when going beyond 1 kHz. This odd, “shaped” response may not be a result of the iPhone 11’s mic sensitivity, however.

Areas of Improvement

Given the environmental interference, the spectral graphs of my recordings are not a true reflection of my iPhone 11’s capabilities. Ideally, this test should be done in a soundproof room. There are also issues with distortion of the laptop speakers. That distortion would be carried into the iPhone and unfairly create the impression that the iPhone has poor performance. Even if Automatic Gain Control can be disabled, the mic remains especially sensitive to the outside environment. The likelihood of background noise interfering with my recordings is very high. If I were to do this experiment again, I would search around for a place that is much more soundproof. 

Information For iPhone Specs

Kunesh, Andrew. “My IPhone Microphone Is Not Working! Here’s The Fix.” Payette Forward, 27 Nov. 2017, www.payetteforward.com/iphone-microphone-not-working-heres-fix/.

“IPhone 11 – Technical Specifications.” Apple, www.apple.com/iphone-11/specs/.

“Smartphone Recording Tips for Recording Birds in the Field.” Help Center, 9 Apr. 2020, support.ebird.org/en/support/solutions/articles/48001064305-smartphone-recording-tips.

What Audio Quality Is Available on Newer IPhones? 2017, apple.stackexchange.com/questions/276905/what-audio-quality-is-available-on-newer-iphones.