Hi, I have a code which captures 8bit speech samples from audio port and dumps it into a file. The codec that I am using uses 16 bit values. I am looking to convert the 8 bit values to 16 bit. I used the following short temp16 = (short)val[i] << 8 ; where val is an array of 8 bit samples. The above statement makes the resultant speech waveform extremely noisy. Can anybody suggest a method for 8bit to 16bit conversion ? Regards KL
8bit to 16bit conversion
Started by ●February 26, 2006
Reply by ●February 26, 20062006-02-26
It's not any noisier, just much louder because of the left shift by 8. Try
just sign extending the samples to 16 bits:
e.g.
if val[i] &0x80
temp16 = 0xff00 | val[i];
else
temp16 = val[i];
"Kunal" <kunal.lagwankar@gmail.com> wrote in message
news:1140958351.945044.266630@i39g2000cwa.googlegroups.com...
> Hi,
>
> I have a code which captures 8bit speech samples from audio port and
> dumps it into a file. The codec that I am using uses 16 bit values. I
> am looking to convert the 8 bit values to 16 bit. I used the following
>
> short temp16 = (short)val[i] << 8 ;
>
> where val is an array of 8 bit samples.
>
> The above statement makes the resultant speech waveform extremely
> noisy. Can anybody suggest a method for 8bit to 16bit conversion ?
>
> Regards
>
> KL
>
Reply by ●February 26, 20062006-02-26
Anonymous wrote:> It's not any noisier, just much louder because of the left shift by 8. Try > just sign extending the samples to 16 bits: > > e.g. > > if val[i] &0x80 > temp16 = 0xff00 | val[i]; > else > temp16 = val[i]; >8-bit codecs usually (AFAIK, always) output unsigned values (range 0 to 255), whereas 16bit samples are always signed (2s complement): range -32768 to 32767. So before the left shift, one has to subtract 128 from the 8-bit value: short samp16 = (samp8 - 128) << 8; Rendering it one may be more aware of the 8-bit truncation noise, but there will not be any ~added~ noise. Richard Dobson
Reply by ●February 26, 20062006-02-26
On Sun, 26 Feb 2006 15:14:50 GMT, Richard Dobson <richarddobson@blueyonder.co.uk> wrote:>Anonymous wrote: >> It's not any noisier, just much louder because of the left shift by 8. Try >> just sign extending the samples to 16 bits: >> >> e.g. >> >> if val[i] &0x80 >> temp16 = 0xff00 | val[i]; >> else >> temp16 = val[i]; >> > >8-bit codecs usually (AFAIK, always) output unsigned values (range 0 to 255), >whereas 16bit samples are always signed (2s complement): range -32768 to 32767. >So before the left shift, one has to subtract 128 from the 8-bit value: > > short samp16 = (samp8 - 128) << 8; > >Rendering it one may be more aware of the 8-bit truncation noise, but there will >not be any ~added~ noise.Some 8-bit codecs will output A-law or Mu-law values. The OP should clarify what "8bit speech samples from audio port" actually means. Regards, Allan
Reply by ●February 26, 20062006-02-26
Hi The samples being read from the audio port are raw values. I am dumping the 8-bit values into a file. The file when opened using Cooledit plays properly. I dont think the values read from the audio port are compressed. I will try out the things given here. Regards KL
Reply by ●February 26, 20062006-02-26
Kunal wrote:> Hi > > The samples being read from the audio port are raw values. I am dumping > the 8-bit values into a file. The file when opened using Cooledit plays > properly. > I dont think the values read from the audio port are compressed. > > I will try out the things given here. > > Regards > KLHi KL if you are using matlab's sound function to playback the audio then you should normalize the shifted signal in the range of -1.0 <= x <= 1.0. It will be played with out any distortion. Regards, gold
Reply by ●February 28, 20062006-02-28
Richard Dobson wrote:> > 8-bit codecs usually (AFAIK, always) output unsigned values (range 0 to 255),Sorry Richard, but thats not right. When 8 bit PCM samples are stored in WAV files, they are stored as unsigned 8 bit values as you suggest. However, AIFF, AU and other file formats store signed 8 bit values (-128 to 127). Erik -- +-----------------------------------------------------------+ Erik de Castro Lopo +-----------------------------------------------------------+ Virgins? What virgins? Its 72 raisins! http://www.guardian.co.uk/religion/Story/0,2763,631357,00.html
Reply by ●February 28, 20062006-02-28
Erik de Castro Lopo wrote:> Richard Dobson wrote: > >>8-bit codecs usually (AFAIK, always) output unsigned values (range 0 to 255), > > > Sorry Richard, but thats not right. When 8 bit PCM samples are stored > in WAV files, they are stored as unsigned 8 bit values as you suggest. > However, AIFF, AU and other file formats store signed 8 bit values > (-128 to 127). > > ErikYes, indeed, file formats can be unipolar or bipolar. I was thinking more of the hardware, since the OP refers to a "codec", which I still tend to think of as hardware more than software. Most of the 8bit audio-rate ADC chips I remember looking at were unipolar; some offered the option of both unipolar and bipolar conversion. But I got the firm impression that for a bog-standard successive-approx ADC, the "normal" configuration was unipolar (e.g. for simple single-rail powering). Maybe modern chips are different? I last looked seriously at 8bit ADC designs at least a decade ago! Richard Dobson
Reply by ●February 28, 20062006-02-28
Richard Dobson wrote:> Erik de Castro Lopo wrote: > >> Richard Dobson wrote: >> >>> 8-bit codecs usually (AFAIK, always) output unsigned values (range 0 >>> to 255), >> >> >> >> Sorry Richard, but thats not right. When 8 bit PCM samples are stored >> in WAV files, they are stored as unsigned 8 bit values as you suggest. >> However, AIFF, AU and other file formats store signed 8 bit values >> (-128 to 127). >> >> Erik > > > Yes, indeed, file formats can be unipolar or bipolar. I was thinking > more of the hardware, since the OP refers to a "codec", which I still > tend to think of as hardware more than software. Most of the 8bit > audio-rate ADC chips I remember looking at were unipolar; some offered > the option of both unipolar and bipolar conversion. But I got the firm > impression that for a bog-standard successive-approx ADC, the "normal" > configuration was unipolar (e.g. for simple single-rail powering). Maybe > modern chips are different? I last looked seriously at 8bit ADC designs > at least a decade ago!The only difference between a unipolar and an offset-binary ADC is a half-scale offset to the analog input. The only difference between an offset-binary and a twos-complement ADC is inversion of the MSB. Offset is in any event needed to deal with bipolar audio. The polarity of the MSB is a trivial matter. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●March 3, 20062006-03-03
Hi all, I have successfully converted 8-bit samples to 16-bit values for my 16-bit codec. However I am facing another problem now. The 16-bit values that I read from the port are very large. Even at times when no speech exists (and only noise exists), the sample values are very large. I found some standard speech test sequences on the net which had very small values as compared to the 16-bit samples file I have generated. To add to my problems, I found that I get very less compression for the 16-bit samples file generated by me and the compression for the standard test sequence is very large. I tried scaling the values down but it didnt help. I am using G.729AB as the codec. Regards KL
