Another improvement for making your workflow more efficient is the new Adobe Media Encoder. This allows you to add multiple timelines or multiple compression ratios and render them all back to back, just like the render queue in After Effects. And, because the Adobe Media Encoder is a standalone program, you can continue to work in Premiere Pro after you drag timelines into the Exporter.
Now, that we’ve gone through most of the improvements, let’s move on to the most advertised new feature for filmmakers: Speech Search. Essentially, Speech Search is designed to process the dialogue in your clips and create an entire XMP text transcript which can be searched dynamically. This searching capability can be accessed in Premiere Pro, Soundbooth, After Effects, and, eventually, Flash, allowing you to quickly find clips that include certain words or clusters of words.
Obviously, this sounds great. However, how did this initial release work in our actual tests?
First off, let’s look at the speed of transcription. It supports the new Adobe Media Encoder, so you can drag your audio clips en masse into this queue and then let it process the audio overnight. The overall ratio that it converts audio to searchable text was averaging about 3:1 (or 3 minutes of processing for every minute of audio) in our tests. However, we did run it through a number of different types of audio files, including music. On clips that were pretty clean, such as decently recorded production narration, transcription was about 2 to 1. On clips that were pretty muddy, such as a rap music clip, it was about 5 to 1. Normal production audio was running approximately 3 to 1.
Now, we move on to accuracy. Unfortunately, this is an area that still needs a lot of work.
On one bit of narration from my first film, Commissioned, I had a line of narration that stated, “Customers are a lot like horror movie victims. They seem to be constantly walking out the door, saying, ‘I’ll be right back,’ but you know you’ll never hear from them again.” What PP CS4 transcribed from this was, “Customers are a lot like war and peace he seemed to be constantly walking in the two were saying we’ll be right back and then you never hear from him King.” Obviously, that transcription needs a bit of work, but it’s not too bad. However, this was the best job Speech Search did on the production audio we tested it with. (I differentiate production audio because if you’re using a better mic to record narration audio, like the USB-equipped Blue Snowball mic, you do get better accuracy than on production audio or on audio recorded on lower quality mics.) Non-narration production audio we tested was at around 25% accuracy while audio that had any sort of background noise, like the rap song we mentioned earlier, completely confused the transcription software. Now, this may make it sound like this new feature is a complete wash, but, in reality, I think it’s simply a new direction that Adobe is exploring that still needs work. (Not unlike the 3D options in Photshop CS3 Extended.) It has the makings of a very useful tool, but it requires further improvement to come to the place where it’s going to be as seamless as it needs to be.
While we’re waiting on new updates for the Speech Search feature to come in, fortunately they do make it pretty easy to touch up words and sub-divide ones that the transcription had problems with. (And, because the narration is tied to the original audio track with XMP timecode data, once you make the corrections, the words are lined up with the spoken words pretty nicely. This makes it much easier to create subtitles after the fact.) Hopefully the CS5 version of this concept will not only feature improved accuracy but will also allow the converted text to be used inside of Soundbooth as looping text for ADR sessions.
Now, perhaps three of the coolest features of Premiere Pro CS4 were actually launched as free updates shortly after the CS4: Production Premium package began to ship and these are all designed to cut into FCP’s market dominance as an editor. (While FCS is nowhere near as strong a package as CS4:PP, the actual FCP editing software is still dominant in market share to Premiere Pro.)
The first is the ability to import newer versions of Final Cut XML EDLs. (Be aware that Native FCP EDLs and older XML files aren’t supported. At least, not yet.) The overall operation works somewhat like Automatic Duck, which imports FCP XML timeline files directly into After Effects. However, with Automatic Duck, if you were just trying to move from an Apple environment into an Adobe environment, it was a pain to have to make visual editing adjustments in After Effects, considering that After Effects really isn’t an editing application. (Automatic Duck was created to essentially give FCP users a sort of “Dynamic Link” with After Effects for color correction and other work, which was especially important before Apple purchased Synthetic Aperture and released Color.) The ability to bring FCP work directly into Premiere Pro means that folks will find it much easier to make any editing changes they may need after they’ve left the FCS workflow and entered the Production Premium workflow.
