Transcribing audio (wav) interviews to text - how?

[Ozzie Alfonso]

It's been ages since I've been around here, but I need suggestions. I am currently doing a series of audio interviews. The interviews can go from 20 minutes to an hour, and then they all need extensive audio editing to bring the time down to 10, 20 minutes or so. The most efficient way to edit the interviews is off a text file, then implementing the text edits and cleaning the audio. I know there are many apps that can "transcribe" an audio file to text, but I'm looking for one that can do it well from the get go. Although each interview is with one person, the series will contain interviews with many different persons or voices.

I'm using a Mac with Adobe Cloud - Premiere Pro, Audition,etc. Any suggestions will be most appreciated.

Ozzie

[Jim Michael]

I don't quite get your intended workflow - are you planning to wind up with audio or text as your work product? If audio, how does transcription or speech recognition help achieve that goal?

I've been working with speech recognition technologies for over 10 years. It's unlikely you will achieve any time (labor) savings attempting to use it to transcribe your audio. Hire a transcriptionist. If you need help finding one shoot me a PM and I'll try to connect you with a couple.

[Don Palomaki]

Most voice to text recognition software output I've seen is pretty bad. (Some of the better laughs I've had at the office come from the Microsoft Lync-Outlook-VOIP combination results.)

But if you intend to use it to select cut points off-line for material with which you are already familiar (e.g., misspelled words are not a problem), it might be good enough if the recorded voices are clear. However, I don't know what is available for the MAC.

[Les Wilson]

Consider that there are so many ways to pronounce words and phrases from one person to the next and then add accents such as New York and non-native speakers and you start to see the problems speech to text technology faces. One technology may be good with one speaker but not the next. The machine transcription won't be 100% and could be as bad as 60% accurate but for your needs, that may be fine.

Transcription accuracy typically is improved by taking a transcription and teaching the software it's mistakes. Also, training the software by having the speaker read a script is another. Looks like neither of those are an option for your workflow so you are left with the default "average" transcription models. I'd recommend looking at a Cloud based service. They'll have refined their models over many hours of audio and may be able to localise for a region like New York.

[Richard Crowley]

My understanding is that virtually all the prime-time shows (60 Minutes, et.al.) use human transcription services where they email the audio file to an organization in another time-zone (typically Asia/Oceania) and it is transcribed by humans overnight and ready for use the next day. If you Google for audio transcription service or similar term, you will find hundreds (maybe thousands?) of such services.

[Dave Farrants]

I've had some success using Audacity / Soundflower / Apple Text.

Info here:http://www.leveluplunch.com/blog/201...y-soundflower/

[Andrew Smith]

Or you could upload your video to Youtube and then download the auto-transcribed .srt captions file. Not sure how this integrates perfectly in to the rest of your workflow, but it's a start.

Andrew

[Chris Medico]

If you need accuracy there is no substitute for using a transcription service. It may cost $1/min but software won't cut it.

[Andrew Smith]

That is true. You will still always need the human layer in the process.

Andrew

[Rick Reineke]

I am frequently asked in ENG type work to record a secondary audio mix for transcription (usually an MP3 and about 25% of the time with a LTC track). All of my clients use professional transcription services... and AFAIK is done by carbon based lifeforms (homosapians..aka, humans). As I understand, the voice recognition software does not work very good for this. But don't that stop you..VMMV, try it and post the results.

[Ozzie Alfonso]

Thank you all for the suggestions. I'll try a few of them and get back. The accuracy of transcription is not important. The transcribed text is being used to determine content and cut points. I'll let you know what I find.

[Jonathan Levin]

With professional transcription services, they also add coughs, uhs, and so on so you know where those are. Well worth the cost!

[Don Palomaki]

For a quick sample of professional real time text recognition - see the caption on live TV broadcasts like FOX News - and laugh (or cry)

[Jim Andrada]

Or visual voice mail on your phone - yuk, yuk, yuk. It always calls me Jim and Rhonda instead of Jim Andrada. Ought to see what it does if my wife (Japanese) tries to leave a voicemail in English - at least it usually just admits total defeat when I get Japanese voice mail.

I do have an acquaintance who's completely paralyzed and dictates books using Dragon Naturally Speaking, but it takes a lot of editing by his assistant even though it's trained to his voice.

[Andrew Smith]

I feel your pain, Rhonda. :-P

Google's voice to text translation is pretty good. The hard part is to convince the input (web browser) that the audio is coming from a microphone source on the computer. It won't budge if it is a line-in source.

Andrew