Speex dev - May 2006 - Transmitting synthetic speech using Speex?

Tom Grandgent wrote:
> Have you tried using 16kHz wideband?  The sound quality is far superior to 
> narrowband, IMO, even if you have to turn the VBR quality down (to say, 2) 
> to save bandwidth.
Thanks for the info Tom!

Probably narrowband is hurting me, but my system is currently built on that. I 
want to try to get acceptable performance from narrowband if possible before 
trying to add wideband support to the system.

Consistant, noise-free synthesized speech certainly has fewer potentially 
problematic aspects to it though, I agree... Maybe we should all just make 
people type their communications in to a keyboard :)

Reed

> 
> Tom
> 
> Reed Hedges <reed@mobilerobots.com> wrote:
>>
>> Hello, I'm wondering if anyone has done this before and has any
advice, or if
>> anyone in general has ideas about it.
>>
>> I just implemented transmitting synthesized speech (text-to-speech)
over Speex
>> (narrowband) in an application.  I'm using Swift from Cepstral 
>> (http://www.cepstral.com). The voice I'm using is a pretty deep
male voice.  I'm
>> telling Swift to generate audio at 8khz, then encoding each chunk of
audio
>> output by Swift and sending it to a client.
>>
>> One interesting thing I've noticed is that as I increase
Speex's encoding
>> quality, the output in the client sounds smoother (at my normal quality
value of
>> 5 or 6 it sounds OK but occasionally has a hesitation or glitch) but
"thinner"
>> -- less full or less resolution.   Using the noise filter and changing
the
>> complexity parameter don't seem to matter.
>>
>> I'll be experimenting with this more, but if anyone is interested I
can send
>> some audio data generated by the Swift synthesizer.  Or, if anyone has
any
>> suggestions for how to tweak the synthesized audio for better encoding
by Speex,
>> that would also be helpful (I don't know very much about audio or
audio signal
>> processing yet.)
>>
>> Thanks!
>>
>> Reed
>

Hi Reed,

I've been using Speex to transmit TTS for years.  It works very well with 
no tweaking.  I use Microsoft TTS ("Microsoft Mike") with Speex at
16kHz
wideband and VBR quality 6.  Sometimes I forget that the sound is even 
coming from another computer and being compressed+decompressed.  If 
anything, TTS seems easier for Speex to deal with than real voice.  But 
I don't have any data to back up this impression.

Have you tried using 16kHz wideband?  The sound quality is far superior to 
narrowband, IMO, even if you have to turn the VBR quality down (to say, 2) 
to save bandwidth.

Tom

Reed Hedges <reed@mobilerobots.com> wrote:
> 
> 
> Hello, I'm wondering if anyone has done this before and has any advice,
or if
> anyone in general has ideas about it.
> 
> I just implemented transmitting synthesized speech (text-to-speech) over
Speex
> (narrowband) in an application.  I'm using Swift from Cepstral 
> (http://www.cepstral.com). The voice I'm using is a pretty deep male
voice.  I'm
> telling Swift to generate audio at 8khz, then encoding each chunk of audio 
> output by Swift and sending it to a client.
> 
> One interesting thing I've noticed is that as I increase Speex's
encoding
> quality, the output in the client sounds smoother (at my normal quality
value of
> 5 or 6 it sounds OK but occasionally has a hesitation or glitch) but
"thinner"
> -- less full or less resolution.   Using the noise filter and changing the 
> complexity parameter don't seem to matter.
> 
> I'll be experimenting with this more, but if anyone is interested I can
send
> some audio data generated by the Swift synthesizer.  Or, if anyone has any 
> suggestions for how to tweak the synthesized audio for better encoding by
Speex,
> that would also be helpful (I don't know very much about audio or audio
signal
> processing yet.)
> 
> Thanks!
> 
> Reed