BenBax

Microphones for Voice Recognition Software
(Dragon Naturally Speaking)

   ADVERT FREE - I'm not selling you anything.   

Microphone Tips

Whilst I am admittedly a microphone fan and collector, and also work as a voice-over artist, my main day job involves writing lots of e-mails each day.
Aside from a headset telephone, colour laser printer and putting your computer monitor in portrait mode if you mostly do emails and documents on your PC, the number one item that will transform the way you do business and hugely increase your productivity is voice recognition software.  I use Nuance Dragon NaturallySpeaking Preferred 10.0 which came with a wired headset microphone. But this headset is rather cheap, ties your head to the computer via a wire, not particularly comfortable and with owning such an arsenal of microphones I never felt I could use compromise with such a 'freebie'.  Newer, updated editions of the software are available and version 10 is obviously no longer in publication.

 UPDATE  - Aug 2023 - After having written all of the below, I upgraded to the latest version of Dragon Professional V16 and have found that now, microphone choice realistically makes little difference! 

For example, there is really no difference in accuracy between my using a Neumann U87 AI costing £2,359, and a Shure SM57 costing £80 - or even a Beringer Ultravoice XM8500 which is about £30, or even a second hand 1986 mic from an online flea market. 

After speaking to a chap in support at Nuance, the 'engine' that drives the recognition in the software now uses a new method to more accurately deduce what words should likely come next in the sentence.  With this new version, I didn't even have to do any training before using it - it just worked straight after installation with an impressively high level of accuracy, even compared with the previous version which I had been training and correcting and using daily for more than 10 years.

The reason I called support was to find out if I could import all of the training from the previous version, but because I'd missed on so many versions in between 10 and 16, the guy in support said it's not possible and because the software is so different now, it's potentially not likely to assist it with accuracy.  The new version even automatically adjusts the microphone input volume level dynamically.

So it appears microphone choice is now not something you really need to get concerned about as much as I did when wirting this article! :0/

However, I still advise a directional cardioid or super cardioid microphone instead of an omnidirectional, and a dynamic as opposed to a condenser will reject more ambient sound from your room/environment and just pick up your voice without any extraneous noise such as clicking or typing or any music you might be listening to in the background.
There are actually benefits to an omnidirectional microphone as you are not in a sound reinforcement/PA situation. No popping, no bass proximity effect, less handling/vibration noise, doesn't have to be pointing in exactly the right direction to you.
See more in this interesting article - https://service.shure.com/s/article/why-should-i-use-an-omnidirectional-microphone?language=en_US

I will keep this article here anyway for posterity and in case it may still be of interest to anyone.

Before I discuss microphones, for voice recognition software the very basics are that your computer should have as much RAM as possible (I have 16 Gb), a fast processor (which will speed up recognition - mine is 3.2 GHz) and I find an SSD solid-state hard drive is such an incredible upgrade to your computer that it can only do good things for voice recognition let alone everything else you do (including searching your hard drive or e-mails if you still store them locally).

I have Windows 8.1.  Negative reviews of the latest version (Dragon NaturallySpeaking 13 HOME edition) suggest that it can be finicky to work if your computer specification is not sufficient, the supplied headset isn't that great, and it may not like Windows 10.  I also don't know if any vociferous anti-virus software interferes with its installation.  It sounds like the PROFESSIONAL edition of v13 does not have these installation problems which after reading a pattern of the reviews sound like they are mostly DVD disc quality related.
I also use old versions of software - Outlook 2007 and Office 2007 ('old' as in 11 years old as this article written in 2018).  Hey, they work absolutely fine - why change and end up paying a monthly fee to use software?  I really don't understand that trend - I would switch to open source versions of software (Linux, Libre Office, etc) before I would ever pay a monthly fee.

Initially it did not dictate into Microsoft Word 2007 until I went into ‘Word Options’, then ‘Add-Ins’ to ‘allow’ it.  This was strange because on other computers with the same software it worked without having to do this.
It does not work in any web browser or any slightly odd software such as Macromedia Dreamweaver.  It will allow you to dictate the filenames for documents can you save them in the save dialogue box, but it will not allow you to dictate into the search box of Windows Explorer to search for a file.

 UPDATE  - in May 2023 I got a new PC with Windows 10, Office 2021, 128 GB of RAM (total overkill, probably should have only got half this), and a 24 core processor. As I determinedly used Dragon 10.0, with it limping along not really working very well in compatibility mode thinking it was running in Windows Vista, I gave up after a month and had to shell out £600 on the current version of the software which does work and is compatible properly. As an aside, I regret buying Office 2021 as Office 2007 would have worked fine and faster I'm told. I've left this article in place for reference. Back to the original…

I also do not know how successfully it works with regional British accents.  Despite being born and bred in the north east, I have trained hard to speak with a neutral accent to appeal to the greatest customer base as possible for my voice over work.  A plumber in Southampton isn't going to want a Teesside accent on his answerphone greeting...
The software works for different languages, but in terms of British accents - a Cockney compared with a Geordie or broad Glaswegian compared with a Welsh accent could surely [and with no disrespect] be heard by a foreigner - or artificial intelligence engine in a piece of software - as a completely different language!?

The plus points of the software once you have it working and set up and trained for a little while (you can get the Accuracy Centre to scan your Word documents or your e-mails for popular phrases and obscure words/names that you use, although I would think this is only if you have files and e-mails locally hosted and not on some sort of cloud or web-based e-mail service) are that even if you are a bad speller, it will spell everything perfectly assuming it is recognised correctly.  You will be able to effectively dictate all your e-mails and letters and any other work into your computer, typing very little each day.  This will be a blessed relief to your hands and fingers, possibly even arms and shoulders and back - surely as humans were never meant to be hunched slaving over computer keyboards staring at screens all day?  Get your office chair set to “recline” and dictate away!

One of the functions in the edition of the software I have is for commands to be set up so you can literally say, “Enter my address” and it will type out your postal address, with the correct line spacing so you never have to type it out again.  Same with anything like a generic reply to an e-mail that you feel you send the same thing to everyone explaining something about your product.  You can train the Command Centre to literally write an e-mail or letter in a sort of template format for you.

Back to microphones! So on this page I offer meandering advice on using different higher quality microphones specifically for voice recognition software.  Although I had used version 7 infrequently before 2005, as my business has grown I've been using v10 intensively since 2010 and my accuracy level through training the software is now at least 99.5%.  I can't tell you how many hours of time it has saved manually typing, but also how much pain it has prevented in my hands, as I used to be a sufferer of RSI from years of typing and mouse work.  As an aside on the topic of RSI, I would highly recommend the 3M Ergonomic Mouse. I also use a Dvorak keyboard layout which also helps. Not drinking carbonated drinks and doing weights excercise to increase blood flow also reduces RSI pain.

The Dragon NaturallySpeaking v10 Preferred software itself cost £93.98 in 2010.  There has never been any software updates since installing despite the software checking periodically.
I have used numerous different microphones with the same software user profile over this time, and because the microphones are such good quality, there has been little, if any, retraining of the software when switching between microphone models. Via the mixer output, the sound signal goes into the same 'line in' socket on the PC soundcard.

Because professional XLR connector type microphones cannot be plugged directly into your computer soundcard, they need to first go through a mixer.  This allows the microphone signal to be amplified sufficiently to be useful for the software to hear your voice properly.
When I first got the software, an analogue mixer was needed.  But now there are many USB compatible mixers which allows you to connect the mixer directly to your computer using a USB port instead of any analogue 3.5mm input jack sockets.  I have one of these mixers for my laptop when I am out of the office and it works just as well.

There are two common types of microphone - dynamic and condenser. 
Dynamic microphones have a smaller pickup area which means they won't pick up any extraneous sounds from the room you are in.
Condenser microphones are much more sensitive and require phantom power to function (which is a low voltage that travels through the cable into the microphone).  Not all mixers have phantom power so you need to look out for this if you are planning to use a condenser microphone, or just want the flexibilty to possibly use it in the future.

Once setup to the correct audio level, you should leave all of the knobs of the mixer as they are at mid-points.  Do not adjust any of the tone controls, or the level once you have started doing any training of the software.  The voice recognition software automatically adjusts the audio level to start with before you do any of the initial training where it listens to your voice to improve accuracy.  As you can imagine, changing the tone of the microphone or the level will mean that the accuracy is greatly reduced because it is not matching your voice to how it was trained. Before starting, I activated any low frequency roll-off or bass cut functions on the mixer, as rumbles from your desk, PC fan, cars going past outside, or a desk fan/air conditioning will not aid accurate recognition.

When you are speaking into the microphone, unlike in a radio or voice over studio, you are not hearing/monitoring yourself back through headphones so you do not necessarily know how good the audio quality is that is being analysed by the software.  So before starting I would recommend you plug in some headphones to your mixer and listen to hear what your voice sounds like at different angles, different proximities to the microphone and bear this in mind when using it from day to day.  You want to make it as easy as possible for the software to recognise your voice, speaking and enunciating words clearly and not slurring or speaking too quickly at different volumes or moving close and away from the mic.  If the software suddenly starts having accuracy problems, the first thing I do is stick a pair of headphones into the socket of the mixer to check everything is okay with the quality of the audio signal being sent from the microphone and mixer to the computer.  Sometimes in the past there has been a buzzing noise or an earth hum from a loose cable which was the reason for the sudden lack of accuracy in recognition.

Always try to keep the microphone the same distance from your mouth, and speak at a similar volume.  I always try to keep the microphone around a hands width from my mouth.  You do not want to have the microphone too close to your mouth as any plosive sounds (when you say words that contain the letter P) will distort the microphone and reduce accuracy. It's also intrusive to have a microphone that close to your face all day. You'll find you knock it with your hand when you scratch your head, etc.

Always try to use the software when you are in a quiet room.  I'm not sure how well it would work in a noisy office - you would certainly need to use a highly directional dynamic microphone to reduce as much as possible any of the extraneous noise from around you such as the MD431 below.


MIXERS

All of the microphones I talk about on this page cannot simply be plugged in to your normal computer soundcard "mic in" socket.  They are professional microphones originally intended for use in recording studios and for radio stations and voice-overs and they require a great deal of gain to provide an audio signal that is sufficiently loud enough to be recognised by the software.  A mixer has the circuitry to provide this gain and also do other useful things like reduce lower rumbling frequencies to aid in clarity in recognition.

“Bog standard” or cheap consumer microphones for Skype-ing your friends or playing on your games console are not as good as microphones used in recording studios or for voice-over work.  As an ex-radio producer and a voice-over artist myself, as well as a prolific voice recognition user, I am fortunate to have the best of both worlds - I use high-quality broadcast standard microphones and audio equipment for my voice recognition dictation.  This gives the software the absolute best quality audio signal so that it gives optimal accuracy and speed. But does not necessarily need to be expensive - an excellent set up can be achieved for around £200. I'm sharing my discoveries with you here.

Try to avoid using the built-in soundcard of your laptop or computer.  This is more important with regards to laptops, as most modern desktop computers do have pretty good sound cards.  Your laptop soundcard however is crammed next to all of the other parts of the laptop including hard drive and power supply so can give a rather noisy signal to the software making it more difficult for it to differentiate between your speech and the noise generated by the computer Plus you'll be forever correcting errors.  In these instances, a USB soundcard is recommended.  Although I do not have personal experience of these, you may be able to plug a professional microphone into the USB soundcard's input socket (via an XLR to 3.5mm jack cable).  However, I personally use a small mixer which I know will provide sufficient gain for the microphone first (as well as other functions such as VU meter to visually observe audio level, low pass filter and phantom power, for example) the line level output of which I would then send to the USB soundcard and then to the computer. A growing number of mixers have USB output connectors, which would negate the need for a USB soundcard at all.

The mixer is quite a small device - it's not some room filling desk with hundreds of knobs and faders like you would see in a recording studio.  The mixers I mentioned generally have one or two microphone inputs and have an analogue output that you can plug into your computer. 

Once initially set-up, you're not ever going to touch the mixer ever again. I had mine on the floor next to the PC for years. The mic will always be 'on', and you'd just turn the voice recognition software to start recognising with the keyboard shortcut (the big '+' on the number pad by default).

I used to use a Behringer Xenyx 502 for years, but it does not have phantom power and I am currently using a condenser mic which needs it so this mixer is currently in semi-retirement.
I currently use an Allen & Heath Zed 6 (shown above), although there are plenty of great mixers with a phantom power function and even with a USB output which will work just as well.


MICROPHONES

This list of microphones below is merely a list of microphones that I have used.  There are literally hundreds of microphone models available, and I'm sure the vast majority of these will work perfectly well depending on the ambient noise level of where you work, and also the acoustics/reverberation (ie an empty echoy room will become a problem compared with a room filled with items to break up reverberation such as carpets, curtains, shelves, etc and no bare walls). Some modern microphones will even have a USB output which allows them to be plugged in directly to your computer without the need of a mixer above. You're really after a cardioid, super-cardioid or hyper-cardioid pick-up pattern, NOT omnidirectional.


Beyerdynamic M201

Type: Dynamic.
Polar Pattern:
Hyper-cardioid.
Frequency Repsonse:
40 - 18,000Hz.

For my initial training and for several years thereafter, I used the M201 microphone.  It cost £112 in 2004 but may be more now.  It's a small microphone, about the size of a cigar, so it has a low profile presence on the desk.  I used it with a large A81WS windshield (designed for the SM57 below) to prevent any 'p' plosives.  With a windshield this large you can effectively lip mike without any issue. This windshield does make it rather a lot larger than a cigar though - more like a cigar stuck in a grapefruit. What an image. You do get a smaller windshield with the mic which is pretty good.

The microphone comes with a microphone stand clip which I used with an 18 inch black gooseneck flexible microphone stand that was clamped to my desk with a G clamp with appropriate thread for the gooseneck.  I kept the microphone cable tidy using cable ties on the gooseneck.  Although standard XLR cable has a straight XLR plug, highly personally prefer the right angled XLR plug and retrofitted the plug to the cable to keep the cable neat and tidy behind the microphone and to reduce its overall length. The gooseneck allows perfect positioning at the right distance and angle wherever you are sitting just by simply adjusting with one hand.

The Beyerdynamic M201 is a dynamic mic, and before the country-wide digital conversion re-fit seemed to be much used by BBC radio stations - mainly BBC Radio 4 (speech based) as a main studio microphone and for panel games / round table discussions and also in all BBC local radio studios. It is still used as guest mics, often mounted on goosenecks in local BBC radio studios (with brightly coloured windshields for easy channel identification). Hypercardioid, small physical size, metal body, and rarely seen for speech without a windshield (supplied), it gives a condenser type sound with rich tones. Because it's a dynamic, you don't need phantom power.

    
The M201 without and with windshield. This is the windshield you get with it, not the A81WS I mention.

M201 Kim Jong Un
(not sure why he needs 5 though - seems a bit greedy)


Sennheiser MD431 II

Type: Dynamic.
Polar Pattern: Super-cardioid.
Frequency Repsonse: 40 - 18,000Hz.

As I confessed earlier, I am a microphone collector so after a few years of using the M201 and getting a little tired of seeing it right in my face every day for years, I decided to switch to a different microphone just for a change.  It is super cardioid which means it's more directional than the SM57 so it does not pick up any noise from anywhere other than the direction it is pointing.  I used this microphone without any windshield - it has a built-in windshield of sorts which is pretty effective at stopping plosives.
I used this microphone on a boom angled arm desk stand so that it was always fairly close to my mouth.  Dynamic microphones are not as sensitive as condenser microphones, and being too far away from this microphone does make a difference to the accuracy level. So a hand span or less really.

Sennheiser MD431 II is Sennheiser's premium dynamic mic with super cardoid pick-up. Sennheiser used the same mic casing for a great number of different models including the Blackfire BF5032P (condenser), BF431 (dynamic), BF531, a range which had specific types of singing in mind (pop, rock, and loud rock which had a moisture resistant capsule!) and a number of radio microphone versions which were the standard seen on Western European TV in the early 1990s. A silver radio mic was even used on Rio's performance at the end of the London 2012 Olympics, much to the amusement of sound engineers around the world on twitter (as they hadn't been manufactured for many years - photo right). The casing design appears to have been used before 1982 - so quite a classic shape. Sennheiser really did stick with these popular mic chassis for decades and decades.

Although originally intended for stage vocals, this model is arguably where it all started for voice recognition software. Even today some websites offer it as a voice recognition microphone due to its directionality. But this all stems from the fact that it was supplied as standard with the IBM Personal Dictation System software back in 1993. In the UK in 1994 the software and hardware including microphone was just under £1,000 and you had to say every word separately in a rather stilted way.

The only description of it in relation to voice recognition now is from a website that sells mics and bundles for this purpose "KnowBrainer.com".
"The same technology that makes the Sennheiser Super Cardioid MD 431 II the world's best vocal microphone also makes it #1 for speech recognition in both noise cancellation and accuracy when speaking within 1 to 4 inches of the microphone element. The combination of the "tailored” frequency response, shock suspended capsule, integrated pop filter and silent on off switch make this microphone unbeatable for "close mic” speech recognition use. Because of its high accuracy, it is able to capture more one syllable words than any other microphone; even when used on the floor of a noisy convention center."
That last bit sounds like first hand experience to me! The term 'noise cancellation' is a bit strong here - the microphone doesn't have any noise cancellation technology or circuitry that sums and rejects signals, but perhaps 'noise rejection' - in other words, just not picking up sound not directly in front of it, may be more appropriate?

But it isn't available everywhere and it's by no means a cheap dynamic mic - in the UK it was £360 in 2007. I got a stonker of a deal from a US dealer in 2007 and paid about £225 for a new one with MZW4032 windshield. If you're using it for voice recognition only, you'd need a USB interface or small USB mixer to get the best sound gain for it to be usable.

I actually use the Black Fire 531 variant of the MD431 for my voice recognition and it is very good and directional.

It has high directionality leading to excellent tone with little proximity effect. I have the MZW 4032 windshield for it with the coloured bands (which is an incredibly tight fit and takes some time and care to get fitted on - despite barely any use, it had perished in sealed storage in less than 16 years after purchase.). It's the first mic I've had in many years which actually has an 'on/off' switch on it! As an aside, it was used as the main on-air microphone on Europe's Radio Luxembourg in the 1980s.

Sennheiser MD431 II
MD431 II laid bare on my mixer (see next page in this article for info on the mixer)

    
MD431 without and with windsheild. Other coloured foam rings are included.

BF531
A different version, the Sennheiser Black Fire 531.
Very similar to the MD531, but with a proud (and detachable) reed switch.

BF531 and SM57
The above mentioned Sennheiser Black Fire 531 in the foreground, with my SM57 with windshield in the background

BF531 suspension mount
Sennheiser Black Fire 531 with cheap windshield mounted onto a Rycote suspension mount to minimise any desk noise from the mic arm.
Note 90 degree XLR plug.


EV ND767a - DISCONTINUED

Type: Dynamic.
Polar Pattern: Super-cardioid.
Frequency Repsonse:
35 - 22,000Hz (?!).

Late one night, I was reading lots of online reviews of microphones and quite a number of people were extolling the virtues of the EV ND767a microphone (what a catchy name...!), saying they preferred it over the SM57. "My goodness!", I thought to myself as an SM57 fan, "can this really be true?". So I bought one. It's certainly a nice mic, sturdy in build with a rubberised soft-touch warm shaft (stop giggling, you). Frequency response is quoted as going up to 22khz, which seems insanely high verging on incorrect for a dynamic mic (but again, is that just EV's different methodology for frequency testing?). It was around the same price as an SM57 but is more directional as it's super cardioid. As it has a larger grille basket this is less prone to 'pops' without a windshield than the SM57, so would be a good choice for voice recognition (no phantom power needed on the mixer) and certainly a very respectable lower cost alternative to the Sennheiser MD531 above.

 UPDATE  I believe it is now discontinued or replaced by something very similar in the new EV range - most likely the ND76 (or the ND76S which has a switch!) or the ND86 which is supercardioid in pickup pattern. The frequency response of these replacement models is now a more realistic 70 - 17,000Hz.

EV ND767a

 


Shure SM57

Shure SM57Type: Dynamic.
Polar Pattern:
Cardioid.
Frequency Repsonse:
40 - 15,000Hz.

Whilst I personally have not used the SM57 much for voice recognition, this is a very popular microphone with good rejection of extraneous noise and there is no reason why it would not make a great voice recognition microphone.  I paid £71 for the microphone in 2006, but they are generally just under £100 today.  You will definitely need to use a windshield as there is no built-in windshield on the microphone.

The Shure SM57 with the huge A81WS windshield (which cost nearly half as much as the mic itself! photo below right) pleasantly surprised me. You see them on every US President's podium since 1965. It is officially an instrument mic used for snare drums and other instruments, and is essentially an SM58 (the classic vocal mic) with a different basket and therefore has a slightly different sound. The SM57 has a very different tone to that of the more 'clinical' condenser mics. I have to say I bought my SM57 more for fun than serious voice work, but it's now my most used microphone for voice-overs.

It will 'pop' a lot from plosive 'p' sounds without a windshield fitted. It's highly A81WSdirectional - rejecting sound from virtually everywhere except directly in front. So for voice recognition you absolutely must use a windshield - ideally the A81WS to eliminate plosives altogether but the smaller A2WS windshield is pretty good and costs less.

The large A81WS windshield has a sort of scouring pad material inside it, presumably to give it better structure and a less muddy sound than if it were made entirely of foam. It's a mid-grey rather than black, fits the SM57 snugly, and eliminates all popping - you will never pop the mic with this on! It's also nice to know in the specs that it is suitable for use when the wind is over 15 mph (which is rare unless I've had a lot of sprouts for dinner).

 UPDATE  - It's well known that the SM57 requires a lot of gain from the preamplifier in your mixer, and you will likely have all of the trim and gain knobs on your mixer maxed out to get a usable level.  The gain is topped out on my mixers.

A new device which wasn't around when I originally wrote this guide many years ago is a “microphone activator”.  This uses the phantom power supplied from your mixer to increase the level of your dynamic microphone so that you don't have to have the gain knob on your mixer anywhere near as high as you may without said microphone activator.
I looked at a lot of reviews, and ultimately decided on the US made Cloud Lifter unit, made in the USA.  There was also a model called a FET Head, and another from SE called the DM1 Dynamite, which looks like a stick of dynamite that plugs into the XLR connector of your dynamic microphone and again uses the phantom power from your mixer (which normally your dynamic microphone would not use or need) to greatly amplify the signal from the microphone.

It does make a difference, and most noticeably on my USB audio interface which really did need the help much more than my mixing desk.
So I would recommend these in addition to any dynamic microphone on this page.  Remember, you need your mixer to have phantom power capability for these to work at all, but it does give you the option to potentially use a phantom powered condenser microphone with your mixer in the future.

A81WS on SM57
SM57 with A81WS windshield fitted

SM57 with A2WS
SM57 with smaller locking A2WS windshield fitted
A81WS podium
The A81WS is a large windshield


A bizarre pairing of my own invention! My SM57 with A2WS windshield and right angled XLR plug mounted 'backwards' in the EV 309A suspension mount designed for the RE20. It does work! It does have a noticably sizeable desk presence, however.


In fairness, a lot of the Shure microphones around this price point sound very similar.  This one above is the Shure 565SD which is made of metal and - quite a rarity in microphones these days - an on/off switch!  For voice recognition I find this is effectively interchangeable with the Shure SM57, Shure SM7B and even a microphone costing more than 7 times as much, the Neumann TLM103.  The key is to be in a fairly quiet environment and keep your mouth the same distance and at the same angle from the microphone as when you trained the software.


The following three mics are really a bit OTT for voice recognition in terms of price/size, but you can sometimes get bargains via online fleamarkets so...

I do like to change microphones every now and then - just for something different to look at every day. However, I usually find myself going back to the Sennheiser MD431.

Sennheiser MD441

Type: Dynamic.
Polar Pattern:
Super-cardioid.
Frequency Repsonse:
30 - 20,000Hz.

Again, after years of using the Black Fire version of the MD431, I fancied another change and noticed my MD441 microphone that I had not really used that I was 'saving for best'.  I purchased this one in 2014 for £279 (second hand, excellent condition, from a Cash Converters shop), although these are normally £700 today.  I purchased the special windshield for this microphone which itself cost £33!
Again, very accurate mic pick-up, nice to look at and used on a boom arm because it's quite a large, long and heavy microphone this one.

The design of this microphone is instantly recognisable worldwide because it's been around and virtually unchanged for 50 + years. Manufactured with only minor cosmetic changes to the design since the 1960s, distinctly rectangular and looking a bit like the front end of an American car from the 50s, the MD441 has been sold with numerous letter suffixes after the name over the years. Few know what they all mean. As far as I can tell, it's mostly to do with whether they have the ring near the XLR plug that rotates to give different bass roll-off settings. There was even a 'Black Fire' model which was all-black and had no bass roll off ring piece which was mass produced as opposed to being 'hand made' like the standard variant. My Sennheiser catalogue from 1994 shows the price as being £404, but currently (in 2014) the RRP is £665. So about inline with inflation.


My MD441 with MZW441 windshield fitted.
Note right angled XLR plug and a gooseneck for another mic at the right.

When I initially plugged it in I was immediately disappointed with the sound quality. Very thin and tinny. What rubbish! Then I realised the bass roll-off ring was set all the way round to 'speech'. I turned it round a few notches to the other extreme - 'music' - and wow - what a lovely sound! Especially for a dynamic. For voice recognition, I actually set it two notches off the full 'music' setting, which is effectively a bit of a bass cut.
You don't see them on TV these days, and when you do it's usually in multiples on some wacky dictator's podium. You know me and windshields, and so I purchased the MZW441 which was horrendously expensive (as all windshields seem to be for older microphones - why?!) and so it's now un-poppable. It was used as the main on-air mic at Europe's Radio Luxembourg in the mid 1970s to early 1980s (with windshield) and I have seen it as the main on-air mic in Japanese radio station studios too.


Illustrating what I've found to be the perfect set-up (for me at least). The MD441 coming from above via a gooseneck mounted on the end of a desk arm, so that the desk arm doesn't get in the way of my reaching to answer the phone, reaching for things on the desk, etc. Windshield used (more in case my chair swivels round and bangs the grille) and 90 degree XLR connector. Bass cut ring piece is 2 notches from 'music' setting, and I also have a low cut filter applied on the mixer.

Dynamic microphones
For scale, l-r: SM57, M201 and MD441. The MD441 is a long mic - about 30cm long with XLR connector.


Sennheiser MKH416

Type: Condenser.
Polar Pattern:
Super cardioid / Shotgun.
Frequency Repsonse:
40 - 20,000Hz.

This microphone cost £700 so I am in no way expecting it to be in the running on anyone's list for voice recognition microphones!  After using some fairly large microphones with a certain 'desk presence' (do bear in mind you stare at these things all day because they are inevitably right in your face on your desk!)  I had an idea to minimise the visual footprint of the microphone I use voice recognition.  This microphone is like a long black stick, and is highly directional as it is a shotgun condenser.  This means it picks up sound from quite a long distance away.  So I can be several feet away from it and it will still pick up what I'm saying.  So it sits a distance away from me on my desk, and has a head-on profile of literally 1.5 centimetres, so very unintrusive.  However, this is the only voice recognition microphone on this list that is a condenser microphone and therefore requires phantom power for the microphone's in-built circuitry to work.  So for this I purchased the Zed6 mixer to replace my Xenyx 502.

I really tortured myself about buying this one as I was going to use it for voice-overs too. So many experienced voice artists and sound engineers say this is not a good mic to use for voice over work. Whilst the MKH416 was never designed for voice-overs studios (it is for TV and film work with a camera, often outdoors), it appears to be the Americans who have persisted with this misnomer and so, converse to its critics, you hear people rave about it too. 

Legendary VO Ernie Anderson can be seen on a YouTube video voicing promos for ABC television in 1989 with one of these inches away from his mouth, no headphones and instead the audio track of the promo he's voicing blaring loudly out of speakers in the same room!  I had seen it as the main on-air mic on BBC Radio 2 in the 1990s (although I think they'd fitted it with a windshield from a much longer shotgun mic to act as a bit of end distance 'buffer') and with royals interviewing each other, below, to reduce room resonance and atmos from the open window, etc.


So purchasing my most expensive microphone (so far) was something I thought about for many months before deciding to do it.  The reason I took the plunge was because of its directionality.  I hoped it would give me the quality of a condenser but without the huge pickup area you'd get with a large diaphragm condenser that would pick up the computer fan and any room reverberation. 

The MKH416 is a long microphone - a little longer than my full hand span, and thin.  Initially I used the windshield with it, but because it is so very directional you can simply talk slightly off axis and you don't need the windshield. 

I bought an elastic suspension mount for it, as it is mounted on a long gooseneck and clamp in my setup. With its directionality, you do have to remain fairly still in the 'sweet spot' of the pick-up. I would say it doesn't 'reject' sound from behind it / the sides as much as an SM57 does, as it's generally more sensitive with being a condenser.

I used the MKH416 for a while in place of the microphone I had used previously for this task - the Sennheiser Black Fire 531, above.  I didn't have to retrain the software switching mics, and the accuracy is very good.  I originally had it on a weighted desk stand with an 18 inch black gooseneck which has an XLR connector built into its end, but because of desk vibrations it's now on a gooseneck coming from a G-clamp stuck to a shelf near my desk.  Because it's a condenser and needs phantom power, and my small Beringer Xenyx 502 mixer I had previously used for voice recognition does not have phantom power, I purchased an Allen & Heath Zed6 mixer specifically for this microphone to use for voice recognition.  I spent a long time researching which mixer to get as I didn't want one with a huge 'wall-wart' mains transformer. It's a very nicely built mixer, and I leave the microphone plugged in and mixer powered up and on 24/7.  It also produces nowhere near as much heat as my Mackie mixer. The only downside is it doesn't have something like a 'tape out' to plug into my PC soundcard (like Mackies do) so I have to take the signal from the headphone output instead as the line level output is obviously too high a signal.

Sennheiser MKH416
My arrangement - the MKH416 on a gooseneck which has an XLR as its end. Very minimal.
The end of the mic is a bit sharp so make sure you're not going to catch your hand on it because it hurts!

MKH416 windshield
A different take - MKH416 with its windshield on, supplied stand adaptor, to a gooseneck.
Note 90 degree XLR plug.


On a gooseneck with suspension mount.


Boundary Microphone

Audio Technica boundary microphone A total spur of the moment purchase from an online flea market website was a boundary microphone because I was buying a couple of other pieces of equipment from the same buyer.
This particular one is a cardioid polar pattern, although many are omnidirectional which would not be appropriate for voice recognition purposes.
It does require phantom power.
You mostly see these microphones for recording police evidence interviews in interview rooms, or on the front of stages in theatres.  You also sometimes see them on the table tops of news studios in case any lapel microphones fail during an interview.
It looks a bit like a computer mouse sitting on the desk, but from over 2 feet away picks up my voice perfectly.  Of course any keyboard activity does tend to disturb recognition, but if you are talking you aren't typing. Room acoustics will also be a factor - probably this won't be too great in echoey rooms with lots of bare surfaces.
When new, these are not cheap microphones, usually around £275.  I got a bit of a bargain for what is effectively a brand-new unit for £35 but very good condition units are around £50 or less, second hand. Understandably, microphones you'll be putting right next to your mouth may not be very hygenic, but you'd be ok with this one as your gob is no where near it.

 


Radio Microphones

TX203In late summer 2018 I had a vision of me pacing up and down in my office whilst dictating e-mails, away from the desk, moving freely and free from cables.  Pacing back and forth is often recognised as a good way of helping you think.  So I got out my Micro TX203 radio microphone, a hand-held mic synonymous with British television in the 1980s and early 1990s (before the smaller Sennheiser SKM4031 and then SKM5000 took over).  It's a large steel microphone with broadcast quality sound transmitting via VHF (for which you no longer require a licence).  With the receiver it uses a total of 3 x PP3 9v batteries, and unfortunately with it being such an old model which I purchased second-hand, I'm not sure on any of the spec, such as how long the battery lasts.  From a Canford Audio catalogue from the late 90s looking at a more modern model TX503, it states the useful battery life is 8 hours.  Again, because the microphone pickup is a condenser of such excellent quality, no software retraining was required.  The only thing I have to do is switch off phantom power from my mixer before plugging it in (it is always best practice to switch off phantom power before plugging in microphones that do not require phantom power).

The microphone capsule pickup pattern of the hand-held transmitter is cardioid and I find extremely directional, so it's ideal.  I have another hand-held radio mike of the same model but it has an omnidirectional pickup pattern as it was previously used for BBC TV newsgathering, so this is not as suitable.  According to an old Canford Audio catalogue from the late 90s, the later model of this hand-held microphone, the TX503, was £875 for VHF, or £1,050 for UHF, excluding VAT.  Then the receiver was £1,350 for VHF or £1,500 for UHF.  Even the audio output cable was £45 and wasn't included. So by no means cheap!


Wireless (Bluetooth) Heatsets

Unfortunately I am one of these people who likes everything to be hardwired.  It even took me quite a long time to come round to WiFi as I simply couldn't get it to work in my house (ultimately due to neighbouring houses' wifi all being on the same channel).  However, there are a great number of Bluetooth headsets available, some of which may just hook over one ear with a boom microphone heading towards your mouth.  With the headset's USB Bluetooth receiver, I'm sure these would provide a good quality audio signal and excellent noise cancellation for voice recognition work.  Perhaps these could even be the solution for noisy office environments? But battery life will be a limiting factor not present with the standard wired microphones above. The main disadvantage may be comfort (I don't like things pressing against my head - plus glasses wearers) and convenience - having to put it on and off just to add a few words to an email, for example.


Lavalier / Tie Clip Microphones

Whilst I currently use the MKH416 which has a low visual footprint on my desk, I have often pondered over the possibility of using a miniature lavalier or tie clip microphone mounted on the end of a gooseneck which would offer truly minimal intrusion on the desk simply due to its minute size.  Or maybe even clipped on your clothes at your chest as a radio microphone? Or just dangling from the ceiling on its own cable? However none of these mics seem particularly low-cost, they are all Phantom powered and the vast majority of those I have seen are all omnidirectional, picking sound up from all around rather than in one particular direction. So whilst aesthetically pleasing due to their tiny size - some of them are about the size of a baked bean - their technical specification may mean they would never really be suitable for voice recognition applications. The much larger Sennheiser MKE40 which is very popular on French TV is a rare cardioid pattern tie clip mic and may be one of the more suitable. Audio Technica also do a cardioid tie clip mic. What I find is you forget you're wearing it, stand up and walk away from your desk, and in the process get pulled back or - worst case scenario - drag a lot of connected equipment with you as you leave!

Something that condenser mics seem to pick up more than dynamics mentioned above is mouse clicks or keyboard button presses. It can lead to mis-recognised words and I've lost count of the number of times I've started having to type in someone's strange name the software would never recognise only to realise I've left the mic open and Outlook has opened the Contacts address book! (a mouse click or a keyboard clunk must sound like "contacts"!).


 

Generally everything works well with the system set up that I have.  As far as I can recall there has only been one major issue over the years and it happened a couple of weeks before I started writing this article.
I had a large whirring external hard drive on my desk that was making a buzzing noise that made the desk vibrate slightly. 
Whilst it was buzzing, I started writing e-mail and realised that the software was having a real job recognising everything and getting most words spectacularly wrong.  So as I mentioned before, the first thing I do is to monitor what the microphone is actually picking up using a pair of headphones plugged into the mixer.  So I unplugged the audio cable that goes from my mixers headphones socket into the audio input socket on the sound card of my computer.  I plugged the headphones in and realised it was the external hard drive making a real din on the mic.  So I moved the hard drive and plugged the cable back in to connect to the mixer with the computer.  But the problem only seemed to get worse - common words and phrases being totally mis-recognised. 
So I went into the accuracy centre of the software to check my audio levels.  To my horror, the software told me that the levels were too low and not loud enough to work properly.  I tried this many times without success.  I turned up the volume to the very maximum which was almost distorted, and it still said the level was too low to work.  What on earth was happening?! It made me wonder whether there was some sort of noise from me unplugging the audio cable which somehow fried the audio input socket circuitry of the soundcard of my computer.  In the end, I ended up having to set up a new "source" for my user profile (so instead of the “microphone/line in” source I had used for all these years, I chose an “array microphone” although I don't think it makes any difference what you call it) and had to go through some general training on the software.  This was bonkers because the exact same microphone and mixer was plugged into the same audio input socket on the same soundcard at the same level and the software was happy with the audio level from the level test to proceed, so I have no idea why the software seemed to think that the level was too low on the source I had been using.  So after using it for 2 weeks, and finding that certain common words were being misrecognised so I was doing the correction for each one to help train it, I tried to get to the bottom of why it stopped working in the first place.  I unplugged the audio signal cable again a couple of times, then opened the old profile with the old input device, and who knows why but it started working again as before.  So everything is back to normal after a week of frustration.  It wouldn't be so bad but I just don't understand why it happened in the first place or what I did to fix it.

Because my soundcard is built into the motherboard of my computer, I started looking at purchasing a separate soundcard like they always used to have so that if anything went wrong I could just replace the soundcard, but then I thought if I did this I would have to retrain the software again because it might see the source audio as coming from a different device being from the new soundcard.

 

 

Neumann U87Microphones - in conclusion...

I don't know anyone else who uses voice recognition software for their business. I'm an evangelist for it, and they are always impressed when I show them how it works. But no-one ever actually purchases it. Maybe too complicated with all the microphone stuff above.

The set-up I would recommend is the M201 or MD431 microphone with a small mixer with USB output. On a lower budget then an SM57 with the smaller A2WS windshield would an excellent buy (again with a USB mixer). As at Jan 2022 this is what I am using. A Cloud Lifter microphone activater will give you more level on any of these mics, but you would need a mixer that provides phantom power to power the Cloud Lifter itself.

The MD441 and MKH416 would really be over the top if all you're doing is voice recognition and the reviews are just published here for interest.

For example, in November 2021 I bought an old discontinued Audio Technica ATM21 dynamic microphone from an online flea market for £40 and in terms of accuracy in recognition I find it interchangable with a Neumann TLM103 that cost £669. Similarly, a collectable EV RE15 mic from 1984, the model used by Elvis at his concerts in the 70s and cost over £400 is just as good as an SM57 costing £80 (and at least with the latter you'd be less precious about getting spit on it!).

Ultimately, if you set up and train your software with one mic and are disciplined to keep it about the same distance from your mouth when speaking, and at the right angle, then any good microphone will work just fine with a suitable audio interface or mixer to pre-amplify the signal from the mic.
I change my mic often and I'm sure that confuses the software much more than if I just stuck with the same model all the time.

 UPDATE  In May 2022 I realised that even your humble Samsung mobile telephone or tablet computer device is equally capable of voice recognition without any prior training or setup using "GBoard".  It doesn't have the features of Dragon NaturallySpeaking with regards to the "Command browser", or correcting, or even capital letters, correcting or training, or lots of little intricacies which make Dragon more usable, but if you need to write a lengthy e-mail from your phone on the go and aren't too worried about accuracy, then I think it could be a viable option.  The downside is Google is involved!

 

Comments, questions, suggestions? Contact me.

Contact me to have your question answered!

 

I hope this helps!

* prices of equipment correct at time of publication.
For price in US dollars ($) simply multiply price you see in £ by about 1.3.

 


 


benbax.com © 2007 - 2025 Ben Baxter. All rights reserved.