Speech Research, AMR modes: speech quality

Hello,

is it true that the 'qualities' of the different modes of the AMR speech
codec not differing very much? In my opinion, the 4.75kbit/s-mode is not
as bad as I expected (compared to the 12.2kbit/s mode). Does a user
perceive a possible switching between the modes in 'real life'? Are
there any published investigations?

Regards,
Marcel

Hello

there are significant differences in speech quality
perceived by listeners. This has been proved by
subjective listening tests and resulting MOS values
(mean opinion scores).

Investigations by H.G. Hirsch prove, that different
AMR modes also have an influence on automatic speech
recognition. MOS values and error-rates are highly
correlated.

< http://dnt.kr.hs-niederrhein.de/ papers/icslp2002_finalfooter.pdf> ;

Greetings

Harald

Have to be VERY careful about switching bit rate modes in
mid conversation. If you choose a bit rate at the start and
hold it all the way through it is better than switching
throughout. Listeners are VERY sensitive to changes, but
quickly adapt even to crappy sounding, but consistent,
quality.

--
Chip Wood



papers/icslp2002_finalfooter.pdf>

of the AMR speech
4.75kbit/s-mode is not
Does a user
life'? Are

Thank you, Harald.

But the fact that the MOS values are relatively high and not far apart
from each other (3.50 and 4.01 for the "best" and the "worst" mode)
proves the overall "good sounding" nature of the codec. Is that safe to
say?

Regards,
Marcel

In article < XXXX@XXXXX.COM >, XXXX@XXXXX.COM says...

yes, in my personal opinion, AMR sounds quite good.
However even a non-experienced listener will notice
the differences between the different mode.

BTW: I am currently working on a web based demonstration system :-)

Greetings Harald

One of the things I dislike about MOS is the way it compresses the
scoring range so much. The difference in sound quality between 3.5 and 4
is huge, even if the numbers might seem close.

Steve