Hello John, Fons, and all who read my post regarding IRs and alleged artifacts. Because my observation was both new (to me) and curious, I did a bit of exploration. If nothing else, it would be important NOT to chop up speech intended for stimuli after applying reverberation. The same could be said for speech recorded in a reverberant environment. John and Fons were (of course) correct in stating that what I hear is the tail of a sound's decay. But in some instances, it's far more pronounced than I would have imagined. If an echo's tail bleeds into a subsequent word, the echo is quite pronounced when one starts from the word's onset. It is particularly noticeable when the sound that created the echo is a broadband sound because it will then sound like the impulse itself. I suppose that's why it's so pronounced. But it really appears loud, and not something that is heard when the wav file starts before any echos are present. There were differences in the onset sound when comparing natural and IR-produced reverberation. With naturally-occuring reverb, a strong "T" sound (a lingua-alveolar stop) will excite room modes and create an audible echo when the wav file is (meaning started slightly beyond the initial production of the T). But it does sound like a "T" sound and not like the "IR" shot that I was hearing. When using speech-weighted noise (600 ms duration, 100 ms rise/fall time) plus a reverb IR, the effect of echoes is quite pronounced when starting playback anywhere in the wav file. Because it's a broadband sound, it does sound like the IR (or a "gunshot"). It is like a ghost in the recording. I next created a pure-tone noise burst (730 Hz--random selection of frequency--100 ms rise/fall time) and applied the same IR used in other samples. Regardless where I started the playback, the result is a pure tone (with echo). There is a noticeable pop if one doesn't start at a zero crossing, but this would be expected. A short rise-time would fix this type of click/pop, but doesn't "fix" processed speech that is started midway in wav file. Just to convince myself that my software doesn't create artifacts, I used an IR of a different type: This time, a stereo HRTF wav file. It sounds quite good, and no pecular sounds or artifacts are present when file is started midway in the sample. Tout va bien. And to investigate other forms of reverb, I took a 1970s recording that used more than a moderate amount of plate reverb. For those of you who remember Neil Young's After the Goldrush performed by Prelude, that was my sample of choice. This was akin to the natural reverb in that clearly-pronounced stops/phonemes can be heard bleeding into subsequent phrases when you begin at a phrase. One likely reason I was hearing so much "gunshot" noise in my original samples is because there was other noise in the recording. The presence of echoes and tails created by the broadband noise gives the "gunshot" sound. None of the artifacts sounded very speech-like, but I assume this is not a fault of the IRs or processing; instead, I assume the underlying noise in the recording is being mathematically operated on when using IRs. Noise simply accentuates the effect. Noise, to include mic self noise, that are not present in the real-world environment will still be operated on by the IR, and echoes of any noise in the wav file become distinctly audible when the wav file is started from any arbitrary point (sans the beginning of the wav file). Lessons learned: I was wrong, but not entirely so. It is clear that recorded speech can't be chopped up and then presented as speech stimuli. The words or sentences to be auralized have to processed as a whole, and then presented to the listener. Even with fade-ins, the effect of lingering echoes is extremely pronounced when the IR comes from an highly reverberant space. It's less noticeable in moderately reverberant spaces, but not subtle. Clearly, arbitrary starting points aren't arbitrary when it comes to creating stimuli. Thanks again for help, and for setting me straight. One way to learn is to experiment and listen carefully. Others already knew what I had discovered for myself, but I think I have a good grasp of what's going on. Listen and learn. Happy Holidays, Eric
________________________________ From: John Abram <johnbab...@gmail.com> To: Eric Carmichel <e...@elcaudio.com>; Surround Sound discussion group <sursound@music.vt.edu> Sent: Saturday, December 15, 2012 6:50 AM Subject: Re: [Sursound] Ghost in Machine Perhaps I'm misunderstanding, but this sounds completely normal to me. The artifacts are simply side effects of starting playback of recorded speech from the middle of a word. Is this situation going to present itself to a person using a hearing aid? I mean does the device itself act as a noise gate? -- with best wishes, John -------------- next part -------------- An HTML attachment was scrubbed... URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20121215/5c2fcd0f/attachment.html> _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound
