Download Audio File, choose format
 Closing Music:Jason Scott Sevin Bits No Waiting
Links: Download Original Interview's Audio (hpr0024.mp3)
Download Remastered Interview's Audio (hpr0024a)
Audacity Project Home Page

Abstract: This is a script for a podcast. Deepgeek discusses how he used Audacity to remaster an old favorite interview he did for Hacker Public Radio, which he always hated the sound quality on.

1 Beginnings of the project

I really got to give thanks to Mr. Klaatu for Bad Apples Episode 4x10 on audio normalization. He has a gift for explaining things, as I did not really understand it before his show. I must say I found it inspiring, as I have an old podcast which is an interview, and was never satisfied with it.

I am specifically talking about Episode 24 of Hacker Public Radio. What was it, I believe my third episode? I don’t even think I knew how to set the levels for recording yet, and here I am interviewing a guy about the IBM Cell BE processor. It is an exotic chip with eight specialized sub-processors called “Synergistic Processing Units.” These SPU’s are geared for matrix processing operations that are used for graphics and heavy math. The interview was good, the sound quality was garbage.

After a few years, the original wav file recorded off a Radio Shack Telephone microphone has been lost to the ebbs and flows of backup disks, but I reasoned that with some years worth of experience with Audacity, as well as my new understanding of audio contrast, that I had a chance, given a little dedication, to issue a remaster (really a re-edit) of the interview. One that would have much better sound quality.

2 Overcoming initial disappointment

Let me describe the original interview, not the content (which I really love,) but the audio.

There is a loud buzz in the background, and my interview subject has a lower audio, as well as a voice that is not as level as mine. My voice has a louder audio level (many would say my voice “carries” or “dominates” conversation.)

Now in Klaatu’s original podcast, he talks about the vast improvement that running audio through a tool called “normalize-audio” can produce. I ran the whole interview through this tool, and got no change in quality.

So, you know how it is, you scratch your head a little, and ask yourself, “what could have gone wrong.” Well, I theorized, that the normalize program was trying to preserve the relative levels of sound for the whole program, that is to say the intro and outro music, my audio levels, as well as Jon Bartlett’s audio levels. Clearly, I realized that this is not what I wanted. What I wanted was the audio to be on level with my voice, Jon’s voice, and the music. That, plus getting rid of that bad background buzz, was what I wanted. At this point, I remembered that Klaatu said that with multiple people, the ideal situation would be to have an individual track for each person.

Well, there was the key, that is when I formulated the plan. I reasoned that an interview is basically people taking turns talking, and that each of these turns can be treated as separate recordings. Therefore, my plan was to take the following steps.

• Remove“uhms, ohs,” and other weird sounds made when people search for words. These sounds, besides being aesthetically unappealing, may have different audio levels.
• Remove my “ahum ahuming” on Jon’s responses, as well as him doing the same to mine. These may have different audio levels.
• Split the audio into two alternating tracks in audacity, to ease processing them as independent chunks of audio.
• For each section, normalize the section, perform a noise reduction, and a “truncate silent spaces.”
• Bring all the parts together tightly.

Now I did this, and it worked. But I must say that it is a labor intensive process. The audio is much more level now, and much less noisy, but it took hours, it was like a regular shift of work. I found this to be very worthwhile, because of the content.

Now, if you are not interested in listening to the whole interview, that is OK. I’m gonna save you the effort as I have prepared a little example.

3 Example

Here is the original clip where I test my understanding of a possible use for this processor, it is 1:03 in length, note the noise, the ums and ohs, and the lower audio of Jon’s voice, as well as his trailing off when he says “simultaneously.” (mp3, ogg)

Here is the second clip, removing ohs and uhms. (mp3, ogg)

Here is the third clip, where I have normalized my voice separately, and then removed noise for myself separately, then I do the same for Jon’s part. I also do a “truncate silence” on both sections, note that Jon saying “simultaneously” gets truncated right at the end. (mp3, ogg)

And finally, here is a fourth example, where I re-normalize sections of each of our parts that have lower audio. In some places you can hear the volume changing mid-sentence. I have not decided whether or not I like this in general, (maybe it would be good for special cases, or maybe on a monologue normalizing each paragraph or section might be feasible.) But it does take more effort as well sounding a little unnatural. However, the last word does not trail off in this version, (you can hear the noise go up with the increase in amplification though.) (mp3, ogg)

4 Conclusion

I certainly would not want to be an audio editor as a full time job! However, I feel that my skills with audio editing are better now than ever before, as well as my now being able to take much more satisfaction with one of my “personal favorites” episodes.