8 Replies Latest reply on Jun 13, 2010 9:09 AM by foochuck

    Speech Recognition - Is It Improved in CS5?

    the_wine_snob Level 9

      In the CS4 forum, a user is trying to work with Speech Recognition, without a lot of luck. The poster asked if that feature was improved in CS5. Any reports?

       

      Thanks,

       

      Hunt

        • 1. Re: Speech Recognition - Is It Improved in CS5?
          Colin Brougham Level 6

          I'm using it with a reasonable amount of success on a current project. I'm not sure if the speech recognition/transcription engine itself is any better than CS4, or if it's the fact that you can now optionally associate a text document or (through a horribly convoluted workflow) an Adobe Story script to guide the transcription, but the result is good enough that it speeds my editing enough to make it a worthwhile tool.

           

          It's far from perfect, but I think the feature is developing nicely to the point where, once it's there, it'll be hard not to use with certain kinds of work. I'm loving it with the project I'm working on, because I can simply mark an in and out around the passage I need to use, and drop it in the timeline--no playback and review required! It might be a lazy way to edit, but it's opening up some options that I might not have tried before, so for that it's certainly worthwhile.

          1 person found this helpful
          • 2. Re: Speech Recognition - Is It Improved in CS5?
            the_wine_snob Level 9

            Colin,

             

            Thank you for the response. A user on the CS4 had some questions, and one was, "is CS5 any better?" I linked to this thread, in hopes that help is right here.

             

            Appreciated,

             

            Hunt

            • 3. Re: Speech Recognition - Is It Improved in CS5?
              Colin Brougham Level 6

              The only thing I'd submit is, "Better how?" As mentioned, it's hard to meter if the part of the software that actually does the transcription is any better, but being able to guide the transcription with a script or even a text document with just some particularly "slangy" words used in the voiceover/interview is a great help. The bottom line is that, no, it's not going to generate a print-ready transcription from the get-go... but it's a decent start.

              1 person found this helpful
              • 4. Re: Speech Recognition - Is It Improved in CS5?
                the_wine_snob Level 9

                As I replied in the other thread, the Speech Recognition aspect only seems to work, if the speaker is Douglas Rain, and his enunciation is spot-on that day. Then, one might get >70% accuracy. Otherwise, the chances seem to fall.

                 

                I was just not sure if CS5 even had retained this, or if it had improved.

                 

                Thank you, and I have linked the OP in CS4 to this discussion, so all that you can add, the better.

                 

                In that CS4 thread, one issue was a non-muxed AV file, with AC3 Audio not being recognized at all. Not sure what the limitations are.

                 

                Hunt

                • 5. Re: Speech Recognition - Is It Improved in CS5?
                  Colin Brougham Level 6

                  I just finished working on a section of my project for which I had no typed script; it was something that was cooked up on the spot while we were shooting the other pre-scripted scenes/sections. In any event, I was not expecting much, but the transcription did really very well--in fact, I'm almost wondering if the transcription worked better than those for which I did have a script. Of course, it did turn "remember, gift card sales" into "dismember kids hurt sales..." Um, no...

                   

                  Re: supported audio formats: I seem to recall that the transcription won't work with MPEG-based audio assets or other similarly compressed audio, like AC3. Not 100% positive on the reason, but I would surmise that it is due to the fact that the transcription actually injects the metadata into the file, and this doesn't work so well with MPEG media. However, I think you can send the audio to Soundbooth, and do the transcription there; it will be linked to the timeline clip in Premiere, in that case.

                  • 6. Re: Speech Recognition - Is It Improved in CS5?
                    foochuck Level 1

                    Colin,

                     

                    I'm working with an AC3 track and I'm trying to get several of them transcribed. Is it possible to send the audio to Soundbooth in CS4 or is that only possible in CS5? I'm not familiar with that workflow, could you point me in the right direction?

                     

                    PS - Thanks Bill.

                     

                    -foochuck

                    • 7. Re: Speech Recognition - Is It Improved in CS5?
                      Colin Brougham Level 6

                      foochuck,

                       

                      This should be possible in CS4, though I can only test at the moment in CS5. Interestingly, Soundbooth won't import an AC3 file, whereas Premiere Pro will (again, this is CS5). But, if you bring the AC3 file into Premiere, you can actually do one of two things to be able to transcribe the audio, assuming you have Soundbooth:

                       

                      1. With the clip selected in a bin, right-click it and select Edit in Adobe Soundbooth > Extract Audio. Premiere will generate a 32-bit WAV (or AIFF, I suppose, if you're using a Mac) and drop it in the bin. It'll also launch Soundbooth with the file front and center.
                      2. Drop the clip in a sequence, right-click it in the sequence, and select Edit in Adobe Soundbooth > Render and Replace. This will basically do the same as above--the audio will be replaced with a 32-bit WAV, opened in Soundbooth, and a clip dropped into the project's bin--but it will also replace the AC3 in the sequence with the newly-generated WAV.

                       

                      From here, you can either transcribe the WAV in Premiere or Soundbooth; the transcription metadata will follow the rendered WAV file wherever it goes. Note, however, that that metadata can't and won't be associated with the original AC3; that shouldn't be a big problem, unless you're expecting it to happen.

                       

                      Hope that helps...

                      • 8. Re: Speech Recognition - Is It Improved in CS5?
                        foochuck Level 1

                        Thanks for the response Colin...I'm going to test this out later today. I'll drop back in to let you know if it works for me.

                         

                        -foochuck