Subject: Re: question about speech level From: Matt Winn <mwinn83@xxxxxxxx> Date: Mon, 23 Jul 2012 13:45:14 -0500 List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>--000e0ce07b1873c1fd04c583a506 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: quoted-printable Ali, A technical detail to consider is that the AC40 typically describes signals in HL rather than SPL. For a broadband speech signal, the calculation is probably trivial for you, but be sure to modify both the target and masker on the same scale if you=92re comparing levels for SNR. Tony=92s suggestion looks like a great way of getting an appropriate SNR, b= ut I would consider more than just leading & trailing pauses. The presence of silences within the sentence is another problem. A recording of a sentence spoken by a talker who pauses a lot (say, =93less-fluent=94) contains less energy in the signal compared to a recording of a more fluent talker. Thus, less noise is required to mask the less-fluent talker to reach the same SNR. This creates an unfair advantage for the less-fluent talker target signal (and on the word level, an unfair advantage for words ending in voiceless stops). I say unfair because the perceived volume of the actual signal content could be the same, but the signals are masked unevenly. Perhaps the ITU Objective Measurement of Active Speech Level will address that problem as well. Matt On Mon, Jul 23, 2012 at 4:45 AM, ali fallah <ali.fallah@xxxxxxxx> wrote: > Dear List > > For creating speech in desired level, I have an AC40 audiometer that > produce speech signal in selected dB SPL from a wav file. > The recordings has silent periods before and after speech signal. If > I delete silent periods before and after the speech signal, the sound > intensity that audimeter produce, will not change. Now I want to prepar= e > a mixture of speech and noise in optional SNR values in Matlab but the RM= S > value of speech signal depends on silent gaps. > > I have two question about this: I want to know for speech signal that is > nonstationary, how desired level in dB SPL is set . Also how can I mix > speech and noise in Matlab by setting noise in desired value (for example > noise level=3D65dBSPL and SNR=3D-10 until we have speech in 55 dBSPL). > I will grateful if anyone guide me about this and let me know about > related documents. > > best regards > --000e0ce07b1873c1fd04c583a506 Content-Type: text/html; charset=windows-1252 Content-Transfer-Encoding: quoted-printable <p class=3D"MsoNormal">Ali,</p> <p class=3D"MsoNormal">A technical detail to consider is that the AC40 typi= cally describes signals in HL rather than SPL. For a broadband speech signal, the calculati= on is probably trivial for you, but be sure to modify both the target and mask= er on the same scale if you=92re comparing levels for SNR. <br></p><p class=3D= "MsoNormal"><br></p> <p class=3D"MsoNormal">Tony=92s suggestion looks like a great way of gettin= g an appropriate SNR, but I would consider more than just leading & trailing pauses. The presence of silences within the sentence is another problem. A recording of a sentence spoken by a talker who pauses a lot (say, =93less-f= luent=94) contains less energy in the signal compared to a recording of a more fluent talker. Thus, less noise is required to mask the less-fluent talker to reac= h the same SNR. This creates an unfair advantage for the less-fluent talker target signal (and on the word level, an unfair advantage for words ending = in voiceless stops). I say unfair because the perceived volume of the actual s= ignal content could be the same, but the signals are masked unevenly. Perhaps the= ITU Objective Measurement of Active Speech Level will address that problem as well. </p> <p class=3D"MsoNormal">=A0</p> <p class=3D"MsoNormal">Matt</p> <p class=3D"MsoNormal">=A0</p> <p class=3D"MsoNormal">=A0</p> <br><br><div class=3D"gmail_quote">On Mon, Jul 23, 2012 at 4:45 AM, ali fal= lah <span dir=3D"ltr"><<a href=3D"mailto:ali.fallah@xxxxxxxx" target=3D= "_blank">ali.fallah@xxxxxxxx</a>></span> wrote:<br><blockquote class=3D= "gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding= -left:1ex"> <div>Dear List</div> <div>=A0</div> <div>For creating speech in desired level,=A0I have an AC40 audiometer that= produce speech signal=A0in=A0selected dB SPL=A0from a wav file. The=A0reco= rdings=A0has silent periods before and after speech signal. If=A0 I=A0delet= e silent periods before and after the speech signal, the=A0sound intensity= =A0 that audimeter produce, will not change.=A0=A0Now I want to=A0prepare a= mixture of speech and noise in optional SNR values in=A0Matlab but the RMS= value of speech signal=A0depends on =A0silent gaps.=A0</div> <div>=A0</div> <div>I have two question about this: I want to know for speech signal that = is nonstationary, how=A0 desired level in dB SPL is set=A0. Also how=A0can = I mix speech and noise in Matlab by setting noise in desired value (for exa= mple noise level=3D65dBSPL and SNR=3D-10 until we have speech in 55 dBSPL).= =A0 </div> <div>I will grateful if anyone guide me about this and let me know about re= lated documents.</div> <div>=A0</div> <div>best regards</div> </blockquote></div><br> --000e0ce07b1873c1fd04c583a506--