Mastering for Voice Communications

Discussion in 'Mastering' started by dpd, Mar 8, 2009.

  1. dpd

    dpd Active Member

    I was wondering if any of the ME's here have ever done any work for purely voice communications (e.g. prepping a track for playing over a bandwidth-limited channel).

    I am doing some work at my job where I need to process live voice to maximize volume, maximize intelligibility, reduce background noise, EQ, etc., over a channel limited to the 300-5,000 hz range. Dynamic range can be limited to the 8 bit range, I'm guessing, maybe as low as 6 bits.

    Hopefully, some of you have had the need to do something related to this over the years.

    Feel free to contact me if you are interested in discussing this. Thanks!
  2. Codemonkey

    Codemonkey Well-Known Member

    I'm guessing it's a form of telephone system?

    I post sermons online from our church but I don't do anything so crushing as this.
    I find 100-10,000Hz works fine and aids the mp3 conversion (48kbps, for ease of playback). Nothing as harsh as 6-bit though.
  3. dpd

    dpd Active Member

    It's for portable voice-projection. Voice in, amplified voice out; therefore, it's analog. Processing will be via DSP
  4. kheftel

    kheftel Guest

    Wow, 6-bit IS really squashing it. I'd say a lot of compression and boosting the EQ somewhat in the 3k range. But I'm a new guy here and I'm sure you already know all that.
  5. Boswell

    Boswell Moderator Distinguished Member

    Is the overriding need to reduce the bit rate? This is a classic instance where companding algorithms can be put to good use. With these, there is an S-shaped mapping of an amplitude value on to a bit pattern (6 or 8 bits). In this way, large amplitudes are represented coarsely and lower amplitudes are represented with a resolution equivalent to 12 or 16 bits.

    There is scarcely a need for mastering as such for this application. Telephone-style filtering using multi-pole roll-offs at the specified corner frequencies will give you the basic filtered PCM samples. Follow this with some AGC or compression to even out the levels and probably a hard limiter for dealing with plosives, and then any bit-compression algorithm as described above.
  6. RemyRAD

    RemyRAD Well-Known Member

    If you can swing this? I find that 16-bit at 11kHz sampling, mono, to be quite nice for the human voice. Response to 20 hertz with a top end no higher than 5kHz. Works well over telephones & limited bandwidth mediums. You would probably do well with 12 bit but I think 8-bit & 6-bit is pushing it from a quality standpoint. Even with companding, it'll get so gravelly. Ugh.

    Go for the smoothness. We want velvety not itchy.
    Ms. Remy Ann David
  7. dpd

    dpd Active Member

    No overriding requirement to reduce the bit rate. processing will likely be done at 16 bit, but the dynamic range will probably be compressed into 8 bits or so to maximize loudness. Intelligibility is key.

Share This Page