[vorbis-dev] encoding lots of speech

[Kevin Marks]

At 8:03 PM +0200 7/14/01, Daniel Resare wrote:
>I had lunch with an interesting guy who had gotten .com-money to record
>the whole bible professionally with good actors (in swedish and english, kjv).
>The idea was to sell custom made compilations of biblical texts on cd over
>internet. The company is now out of money (surprise!), but all the material
>is recorded (about 350 hours) and if anyone gets a good idea on what to use
>the material to, it might be possible to get permission to use it.
>
>Now to my question: how well fitted is ogg/vorbis for compressing high
>quality speech? I've done some limited testing with beta4 that indicates
>that bitrates around 64kbit/s (mono) is needed to make it sound ok.
>How much 'room for improvement' does the vorbis format have on that number?
>Would it be possible to hack a special decoder that was better at encoding
>and how much work would that take?

Speech and music are considered different problem classes by many 
audio codec developers. If the recording is just clear speech (ie not 
mixed with music intros and effects) then there are several codecs 
that will compress it much smaller than 64kbps and still sound OK. 
The GSM family is one, PureVoice (included free in QuickTime) is 
another, and MPEG 4 defines some new ones.

With speech, the harmonic content is less important, and there is 
generally lots of silence that can be removed, but there is also a 
great deal of shaped noise in the form of sibilants (s, f, h sounds). 
The ear is quite sensitive to distinctions in the noise, so reducing 
bitrate by cutting down samplerate can sound wrong (sss turns into 
shhh if the HF is cut off, for example).

My understanding of Vorbis is that is a general purpose compressor, 
and so one could make codebooks tuned for speech, but they would 
require some work.

Based on earlier discussion here, 'noise gating' your speech first 
(putting it through a filter so that there is complete digital 
silence between words) would help Vorbis make it a lot smaller, as 
otehrwise Vorbis will carefully preserve the nuances of your 
background noise in the recording studio well below the threshold of 
hearing.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.