Registering and logging in removes this ad.
Registering and logging in removes this ad.
Want to Use Speech Recognition for Interviews
Submitted by LanaNelson on Fri, 01/25/2008 - 19:33.
I am involved in a community history project to record and transcribe memories of older residents. I have done one of these and it took me about five hours to transcribe a little over an hour of questions and answers. I am investigating speech recognition to see if I can use a program to get most of the conversation to text. I have to convert these recordings into "stories" so they will require heavy editing and rearranging after they are transcribed so perfection is not necessary. I discovered that my Sony VAIO laptop with Vista Home Premium has speech recognition and a built-in microphone. I have been playing with it for a couple of days and .... to say it is "bad" would be a complement.
Can anyone tell me if there is a program that might meet my needs: I will be interviewing multiple people. It is not a problem to have each person read something for 10 or 15 minutes so that the program can hear them speak. I need to be able to save each person's info in case I should need to do later interviews with them. The person won't be doing any punctuation so the program would just be hearing a stream of speech. I will be doing a sound recording ( I need to hear the cadence and rhythm of the speaker's voice) so I can edit for punctuation and miss-interpreted words. Are there are any speech recognition programs that might work so me? Any input will be appreciated.
Thanks,
Lana Nelson
The Vista speech recognition
The Vista speech recognition program has an excellent reputation. OTOH, the on-board microphone does not.
People have often asked this question, so searching the archives might be useful. What is different in this case is the suggestion that you can afford to give people 10-15 minutes initial reading time. Whether this would make it worth while to use a speech recognition program is beyond me to answer, but it might work enough to lessen your workload.
The traditional suggestion is to have someone re-dictate the interviews while listening to them through earphones. With the right kind of gear and a little practice this seems to be feasible.
Echo Dictate
The Vista speech engine probably isn't quite up to meeting your
needs but perhaps the biggest reason why your transcription came out ugly is
because you were using the integrated soundcard
& microphone. This is an absolute no-no. Our Microphone
Comparison Matrix includes a list KnowBrainer approved recommendations. Speech recognition can only be as strong as
its weakest link.
The only way you can transcribe each individual end-user, in an
interview situation, would be if they had separate computers with separate
user files but putting together the interview at a later date could be a
rather daunting task. You might
consider using a decent digital conference recorder, such as the Olympus DS-50,
and just echo dictating the transcription into NaturallySpeaking Preferred
9.5 (which would be our dictation software recommendation for this
situation). You won’t be able to
transcribe the recordings but Echo Dictating is almost as fast as recording
the dictation. You basically play back
the recording, wait for about 1.5 seconds and nonstop echo dictate everything
you hear into DNS.
Additional responses to this
question are available on the KnowBrainer
Technical Support Forum
Lunis
Orcutt - Developer of KnowBrainer &
Host
of the KnowBrainer Speech Recognition Forum
A Nuance Gold Certified Endorsed Vendor
ALWAYS Ask If Your Speech Recognition Vendor Is
Nuance Certified
Interviews can't easily be
Interviews can't easily be transcribed using voice recognition software. No software for PC's can discern between voices. Additionally, people in interviews often talk over each other, they don't dictate punctuation, and they don't talk in complete sentences or phrases.
The only solution we have had customers use is to listen to the interview and re-dictate into Dragon NaturallySpeaking using their own voice. That way, they can also add punctuation.
Bret Williams
www.NovuScript.com
Re: Wanting to Use Speech Recognition for Interviews
Thanks to all for replying.
BruceCyr, I wanted to check the archives before posting but I didn't find any way to do it. If there is some link on the page that will allow me to search the archives, please let me know because I did not recognize any as such. After my post appeared, a area to the right listed "Similar Entries" and you are right, it was most valuable but I wouldn't know how to find it again if I needed to view it at a later date.
Lunis, you and Bruce recommended that I buy a high quality microphone. When I went to your link to microphones in your reply, it sends me to a page last updated in May 2005. Is that still your recommendation? Also... Do I have to get a new sound card?
Bret, I know that speech recognition does not lend itself to interview situations but I was hoping that my particular situation might do better than others. These will be oral history interviews. Only one person at a time will be interviewed. I hope to be able to ask short questions such as "What do you remember about the Depression?" and then have the person talk for 10 or 15 minutes and then I will ask another question. I could even write questions on a piece of paper and have the person read the question and then answer it. In these interview situations, before we start, even the day before if need be, I can have the person read something for 10 or 15 minutes to train the speech software, but they will just talk for the interview -- they would not be supplying any punctuation and I realize that is a problem. In looking through the Help for Vista Speech Recognition, it looks to me that I would have to set up another computer user with separate ID and password in order to establish a different voice profile. Do you know if that is the case?
Another question: Chuck Runquist has a very interesting post in the archives about loading a .wav file directly into Dragon NaturallySpeaking. He says it must be a .wav file in a specific format range. Has anything changed since his post 2+ years ago? The information as to format and sampling was unfamiliar to me so I don't have any idea as to what on the market creates what. I have a digital recorder but it creates mp3 files which would not be usable according to his post. Chuck also says that Dragon allows for the creation of multiple profiles, unlike Vista Speech, which I am assuming would mean that I could have my interview subjects create profiles which I could give their names then I could recall those individual profiles at will if I had to conduct a second interview. I also don't see any instructions in Vista Speech Recognition Help for loading a sound file so I am assuming it can't be done.
Thanks,
Lana
LanaNelson wrote: I wanted
I wanted to check the archives before posting but I didn't find any way to do it. If there is some link on the page that will allow me to search the archives, please let me know because I did not recognize any as such.
There's a blank area in the upper right corner of every page. It says "Google Custom Search" within it with a button labeled "Search" to the right of it.
If you put your search criteria in there (it uses the same commands as Google) it will search many speech recognition sites and return the results on a new page.
If you want to limit the search to ONLY Speech Computing, then you click on "Only Speech Computing" under "Refine results for" on the left side. It will then narrow the search results to only those from this site.
It's updated quite frequently by Google.
Hope that helps!
Skip
Quote: Lunis, you and
Lunis, you and Bruce recommended that I buy a
high quality microphone. When I went to your link to microphones in your
reply, it sends me to a page last updated in May 2005. Is that still your
recommendation? Also... Do I have to get a new sound card?
We are not sure where you got the date from but that was
probably for the original Microphone
Comparison Matrix . It was last
updated about 2 months ago but the Hybrid
Plantronics CS70N wireless telephone/microphone and Hybrid
Senneheiser BW 900 Bluetooth wireless microphone still need to be added
to the list. However, you can also compare
those microphones by clicking directly on the previous links.
Speaking of links… Any weak links in the speech recognition
chain can cause your accuracy to suffer.
Most integrated soundcards tend to be a bit less than ideal so we usually
recommend a USB
Pod (external soundcard) except in situations where a good PCI sound card
is used such as a Sound Blaster or Turtle Beach.
Chuck Runquist has a very interesting post in the
archives about loading a .wav file directly into Dragon NaturallySpeaking. He
says it must be a .wav file in a specific format range. Has anything changed
since his post 2+ years ago?
Before Ver. 9, DNS was only capable of transcribing 11.025 kHz wav
files. Ver. 9 can additionally
transcribe .WMA, .SR, MP3 and DSS files
Lunis
Orcutt - Developer of KnowBrainer &
Host
of the KnowBrainer Speech Recognition Forum
A Nuance Gold Certified Endorsed Vendor
ALWAYS Ask If Your Speech Recognition Vendor Is
Nuance Certified
"but I was hoping that my
"but I was hoping that my particular situation might do better than others. These will be oral history interviews."
Lana, I think if you make the respondents read one of the training passages and also try to get them to enunciate clearly, you run the risk of making them self-conscious (even more so than they would be otherwise!) and so, affecting their accounts. I agree with the people who say you should 'echo dictate' (listen to the interview and dictate into NaturallySpeaking/Vista). You may anyway need a better sound card and mike, I don't know.
But if you want to try the 'transcribe from recording' route, really you do need to get a recorder that's suitable for this kind of work (I don't know which one you're using now), even though you can of course change the file format from mp3 and also clean up the recording using, say, CoolEdit.
Judy
Want to Use Speech Recognition... or get a job done?
What is your objective, a, or b?
a) get a job done? What is the overal best way to do so?
b) use speech recognition? If so, no one can stop you.
Geeks may geek and
folks may freak
but a good way to work
it's good to speak
Sorry about the poem (it was inspired) because your approach seems to be throwing high tech on something that a trained professional (human) can complete better than a "b" computer.
If your objective is to get the job done, you may want an "a" objective.
Good luck, and ... write or call if you need "a" help.
Alan Kelly, owner/operator
www.VerbatimIT.com
Verbatim Instant Transcripts
(before AOL had mail)
802-578-3789