From: "Amir Majidimehr (Exchange)" 
To: "'david@weekly.org'" 
Subject: Feedback
Date: Sun, 18 Apr 1999 22:21:56 -0700

Hi David. Congratulations on being the first to produce an independent comparison of MSaudio against other codecs. I work for Microsoft and manage our audio and video compression group. I am travelling this week and could not access all the clips using my 28K dial-up modem. So, I did appreciate the summary write-up. After I get back in the office, I will listen to the rest of the clips and send you a more complete feedback....

One key peice of feedback that I have is that despite your kind remarks about our resampler, we find that it performs quite poorly, especially when used to downsample 44Khz to 22Khz. The "watered down" effect is actually a symptom of this. The resampler tends to add "warble" to voices and reduces low-frequency impact and energy. Notice how your 44Khz samples didn't suffer from these problems. We are planning on fixing this and also adding additional filtering to reduce some of the excess high frequencies when the code becomes final (mid next month). For now, almost all of our customers use Sonic Foundary's SoundForge 4.5C to downsample and batch convert their content. The SF resampler, when used at the highest quality setting, completely eliminates the above distortions in addition to reducing some of the high frequency aliasing caused by our fast, but less than capable resampler.

We are also working on our performance in speach area. Even though we will never be as good as a dedicated speach coder such as ACELP (which as you know, is also shipped as part of our system), we hope to do much better than we do now. For the time being though, we have found that 16K sampling seems to work best. Lower sampling rates cause artifacts of their own despite improved high frequency filtering and higher rates make the sound just too bright.

Keep up the good work,
Amir Majidimehr
Group Manager
Codec Development
Microsoft Corporation