Dragon Naturally Speaking

[laukce01]

Is there an alternative program to Dragon Naturally Speaking (audio to text) for Linux?

I would like to record notes via a digital audio recorder and then have them transcribed into text using a program such as Dragon. However Dragon does not supposidly work with Linux.

[xenopeek]

There is Simon, an open source speech recognition application. It probably integrates best on the KDE desktop environment but should work fine on others also. Here is a blog talking about the dictation plugin, scroll down for a video demonstration of that: http://grasch.net/node/22. I've not used it personally. On Linux Mint 16 you can install the package "simon" from the Software Manager. You'll probably need to add or at least enable the dictation plugin. On earlier version of Linux Mint you had best compile Simon from source as the packages are too old (http://userbase.kde.org/Simon/Developme ... ment#Linux).

[Laugh2]

Dragon Naturally Speaking (DNS) can work well on Linux , at least some older versions can. All you need is Wine and sufficient RAM. (I have ver. 9, which is not the latest version i.e. DNS ver.12, so I cannot comment on postings elsewhere that the latest version has difficulties on Linux.) I am dictating this post using:
- Dragon Naturally Speaking ver. 9 Preferred (DNSv9)
- Wine 1.6.1
- Linux Mint 13 Maya, 64bit, MATE
- 2GB RAM

Limitations
1. Sufficient RAM
It is necessary to have sufficient RAM available when starting DNSv9. On start-up DNSv9 requires something in the order of 1 GB of free, available RAM. Most of this will be freed up after start-up. On my 2 GB system this means that I have to shut my e-mail client (Thunderbird) while starting DNSv9, but can start again once DNSv9 is running. I could get more RAM to the computer, but it is an old laptop so I'm not too bothered.

If you do not have sufficient free RAM available when starting DNSv9 the results are erratic. Frequently CPU usage on one core will rise to 100% and stay there, usually the program will become unresponsive and the System Monitor is necessary to end DNSv9 before a restart. I was using DNSv9 on Windows XP for many years and had exactly the same limitation. I note that the latest version of DNS specifies a minimum of 1 GB for Windows XP and Windows Vista, and 2 GB for Windows 7, Windows 8, and server versions.

2. Function Limits
Basic dictation into DNSv9's own utilities e.g. DragonPad and Dictation Box, works well with good accuracy. Copy and Paste then works well into other programs e.g. LibreOffice.

DNSv9 on Linux can also accustomed itself to your voice patterns and learn new words, however, training for specific words can lead to a program crash. Many of the other functions either don't work or can lead to stability issues, but since I don't use them this doesn't bother me.

Suggestion
As far as I know DNS still delivers best voice recognition accuracy, so you might consider buying an older version of DNS for use on Linux e.g. through eBay.

Details:
Linux Mint 13 Maya (64 bit) MATE
Dell Latitiude D630 (Core2 Duo, nVidia Quadro NVS 135M graphics)

[Lingula]

RAM (min 4GB, better with 8GB) and hard drive space (70GB spare on the drive with Dragon is preferred due to the incredible volume of temporary files) are the key. Specific versions can work under Wine. Just search WineHQ and you'll see. It wasn't stable enough for me to use at work, so that's my one remaining dual boot. Linux native speech apps thus far are mostly for commands rather than transcription.

[XE1UFO]

We have several forum active members who have physical disabilities and are looking for this feature. I am certainly surprised how little is available in proprietary Linux for voice dictation/speech recognition. These are available by the truckload for free on Android, so it should not be too hard for somebody to port to Linux, in my humble opinion.

And on both Android and Ipad, speech recognition works very well.

[Lingula]

There are voice command solutions that work well in Linux, but nothing reliable for transcription. Some have had success with Naturallyspeaking in wine. Unfortunately, with the notoriously buggy nature of Naturallyspeaking, it becomes difficult to troubleshoot a wine installation. Nuance tried to charge a fee (maybe still do) to discuss issues with their tech support people. The Nuance forums aren't much help either, since the moderators censor and delete posts that cast Nuance in a bad light or mention any competitor's product. It is the sole reason for my dual boot setup... Sigh.

Rant over.

[Laugh2]

I added 1GB RAM recovered from a faulty computer to make 3GB (refer my posting above). DNS9 now starts more reliably but occasionally still makes the problems noted above.

With care 3GB and even 2GB can be useful, esp. as accuracy and speed of recognition is in general not related to RAM. (Accuracy and speed are determined by a setting in DNS9 whereby system response becomes very sluggish at high recognition levels on less powerful computers. The version of DNS also makes a difference.)]

Later ...
With 4GB RAM I am getting the same issues as with 3GB, especially if previous activity had caused Virtual RAM on the HDD to be used - even when Physical RAM shows plenty RAM is available. In one case even multiple start attempts didn't help and a complete computer restart was required.

What does help is to wait up to a minute after starting DNS9 in Wine but before choosing the User in the "Open User" window. This often gets DNS9 working. If you are tracking CPU usage in System Monitor then this corresponds to waiting until long after CPU activity appears to have returned to background levels. Apparently something is still happening in which is not visible in CPU History.

In general, the longer I wait after starting DNS9 and before choosing the User the better.

[eolith]

As a seasoned user of Dragon Naturally Speaking on Windows (user since version 1.0) I finally made an effort to make it work on Linux using Wine.

Here is my experience - hope it's of use to anyone else reading this thread.

I pretty much followed the instructions on
http://binaryimpulse.com/2013/01/instal ... ayonlinux/

As this URL indicates, the implementation suggests using PlayOnLinux, which you can install from the repository through the Software Manager.

So I ended up using the following combination of software.

• Mint 17 Xfce
  PlayOnLinux 4.2.2
  Wine 1.7.25
  Dragon NaturallySpeaking 11 (Dutch version, which also has English, French and German)

I'm working with a Plantronics Blackwire C210 headset, which is my favourite headset when working with Dragon NaturallySpeaking on Windows.

Without any issues I managed to install it, and it works just fine. Some error messages that were predicted on the aforementioned site I also observed, but they can be safely ignored.

I'm using an old computer for this kind of testing: a Phillips freevents laptop, with only 1 GB RAM. Still, it is workable even with this high version of Dragon, and the recognition is ok enough for usable dictation - if not much else is running on this computer. Of course, some tweaking can still be done to make it faster (Dragon settings on recognition speed vs accuracy, and better training) but given this success, but I will just install it on a faster computer now I know how to do it. As said before, dictation is only possible inside the DragonPad, which is the native .rtf document editor.

So, there, it works, hope you have succes too. It certainly made my day!

[Habitual]

I am certainly surprised how little is available in proprietary Linux for voice dictation/speech recognition.

I am waiting for the "Next Big Thing" in Linux and I hope it's voice recognition.
I am tired of typing after 20 years in IT.

[eolith]

I hope so too, but I do not see any project reaching maturity soon.

A while ago, I contacted someone close to Nuance (creators of Dragon Naturally Speaking), and asked them whether Nuance is considering developing a native Linux version of Dragon. After all, they have already available a version for Mac OS, so how much more trouble would be to develop it for Linux!?

The answer was negative, because the market is too small. Also, the final product would cost money, and it would obviously not be open source, making the market even smaller for a Linux audience.

(As an aside, my experience with Dragon for Mac is that it is way behind the quality and user-friendliness of the Windows version.)

Still, it would be grand if an open source project was started in this area. I would happily support it, and even give money if it was set up as, e.g. A Kickstarter project.

Until then, we are stuck with Dragon NaturallySpeaking combined with Wine, which, as I mentioned in my previous post, can be get to work pretty well.

[BenTrabetere]

Still, it would be grand if an open source project was started in this area.

Whatever happened to IBM VoiceType that shipped with OS/2 Warp? I know it could be used for dictation and for desktop navigation, and I'm pretty sure a version was available for Windows. I wonder if IBM would be interested in a Linux project?

[Mark Phelps]

I wonder if IBM would be interested in a Linux project?

Maybe ... but, probably not. IBM, like the other huge corporations, is in business to make money -- for itself and for its stockholders. Giving stuff away for free is not in keeping with that business model. Open-source really doesn't even enter the picture; instead, it's an issue of profits -- making them and growing them. And, you can't do that with a community of folks that are opposed, in principle, to paying for software.

[DrHu]

Its kind of sparse, but here's looksee..
http://www.brighthub.com/computing/linu ... 92561.aspx

• Xvoice: Xvoice can be used both for dictation and for voice commands. Xvoice is free and open source, but in order to function properly, it needs ViaVoice form IBM (which is not free).

http://www-01.ibm.com/software/pervasive/viavoice.html
http://www.nuance.com/index.htm
Nuance, and this is their link (Dragon Natiurally Speaking..)

http://www.junauza.com/2010/03/text-to- ... linux.html
https://wiki.archlinux.org/index.php/Festival
festival was well known in the past; not sure how well it has faired
--the link has some info on that program..

Dictation might be more of a problem, but text-to-speech is available.
--may or may not be optimal, if you were used to Dragon naturally speaking or ViaVoice..

[Laugh2]

Well, it's been a bit more than a year since the last posting. Since my postings above I have installed LM17.2 with a fresh installation of DNS9 using PlayOnLinux (using WINE 1.5.31). DNS9 now starts every time without a hitch and runs smoothly.

I used DNS9 a lot on Windows XP a few years ago. I almost exclusively dictated into DragonPad because on XP everything else was either too slow, too unstable i.e. caused crashes, or too buggy. So I don't miss those other functions running in Linux (where they tend not to work at all). But as I said, I didn't use them running Windows either.

My subjective impression is that DNS9 is running on Linux Mint with PlayOnLinux better than XP. But since this laptop is faster than my old laptop and I can run DNS9 with maximum recognition accuracy then the comparison is a bit unfair. It certainly isn't slower though.

So now I use DNS9 again for longer passages of text rather than typing, and enjoying it.

Equipment

• Linux Mint 17.2 Rafaela MATE (64 bit)
  PlayOnLinux 4.2.9 using WINE 1.5.31
  Dragon Naturally Speaking 9 Preferred
  Dell Latitude D630 Laptop (Core2 Duo)
  Logitech AK5370 USB microphone