I'm a member of the team from Rice University that went up against the guys from the Technion in the DSP Challenge finals. You can whine all you want about wheter audio watermarking is the right thing to do, but you can't deny that these guys put together a *really* cool system.
Their presentation to the judges was very impressive (we presented first, and all I could think while watching theirs was "we're screwed..."). They demonstrated both the addition and detection of a watermark for pre-recorded and live audio. A couple of times, they played out loud just the watermark. It was pretty garbled, but you could definitely make out the content of the original audio. They did it for both vocal and instrumental music; both times, you could make out the lyrics and melody of the original sound. And when combined with the original, the watermark was inaudible (as promised).
But the most impressive part of all is that it was all done in real-time. They watermarked the audio from a local radio station as it was being broadcast, playing both the resulting watermark and watermarked audio with virtually no delay.
I told them in Dallas, but I'll say it again- Congratulations on winning. You guys were definitely worthy competition. With any luck, we'll face each other again next time around. -patrick
Re:Not that impressive
on
DSLBlaster?
·
· Score: 1
Not quite...
Shannon had a couple of good theorems, but the one you're talking about is courtesy of Nyquist. To have any chance of recovering sampled data, the sampling rate must be at least twice the highest frequency content of the original, analog signal. The original poster's calculation (16 bits/sample * 48 ksamples/sec) was right, but it assumes a noiseless channel.
Shannon's noisy channel theorem is probably more applicable. (the heorem's result is likely what he meant by "Shannon's limit.") Shannon said that a channel's capacity in bps equals channel bandwidth(Hz) * log2(SNR+1). A classic example is the plain old telephone system. The bandwidth allowed is about 3kHz (300Hz - 3.3kHz). The SNR the telcos try to maintain is around 35db, or about 3100. This gives a maximum reliable capacity of around 35kbps.
One fun side note: the 3khz bandwidth limit is not an inherant limitation of the copper telephone lines. It's actually the frequency range chosen as sufficient for voice communication by none other than Alexander Graham Bell. His original equipment used this 3khz bandwidth, and all subsequent telephone equipment adhered to the same standard. Though it was made more than 150 years ago, his choice for bandwidth stuck (and is the reason many of us *still* can't get DSL connections to our homes!).
Slashdot Mirror
Their presentation to the judges was very impressive (we presented first, and all I could think while watching theirs was "we're screwed..."). They demonstrated both the addition and detection of a watermark for pre-recorded and live audio. A couple of times, they played out loud just the watermark. It was pretty garbled, but you could definitely make out the content of the original audio. They did it for both vocal and instrumental music; both times, you could make out the lyrics and melody of the original sound. And when combined with the original, the watermark was inaudible (as promised).
But the most impressive part of all is that it was all done in real-time. They watermarked the audio from a local radio station as it was being broadcast, playing both the resulting watermark and watermarked audio with virtually no delay.
I told them in Dallas, but I'll say it again- Congratulations on winning. You guys were definitely worthy competition. With any luck, we'll face each other again next time around.
-patrick
Shannon had a couple of good theorems, but the one you're talking about is courtesy of Nyquist. To have any chance of recovering sampled data, the sampling rate must be at least twice the highest frequency content of the original, analog signal. The original poster's calculation (16 bits/sample * 48 ksamples/sec) was right, but it assumes a noiseless channel.
Shannon's noisy channel theorem is probably more applicable. (the heorem's result is likely what he meant by "Shannon's limit.") Shannon said that a channel's capacity in bps equals channel bandwidth(Hz) * log2(SNR+1). A classic example is the plain old telephone system. The bandwidth allowed is about 3kHz (300Hz - 3.3kHz). The SNR the telcos try to maintain is around 35db, or about 3100. This gives a maximum reliable capacity of around 35kbps.
One fun side note: the 3khz bandwidth limit is not an inherant limitation of the copper telephone lines. It's actually the frequency range chosen as sufficient for voice communication by none other than Alexander Graham Bell. His original equipment used this 3khz bandwidth, and all subsequent telephone equipment adhered to the same standard. Though it was made more than 150 years ago, his choice for bandwidth stuck (and is the reason many of us *still* can't get DSL connections to our homes!).