VoIP Mailing List Archives

[saeedahmad1981 at gmai...]

Hi all,
 
Could you guys recommend me any online text to speech IVR software which works OK with FS. i am using AT&T site and for some IVRs i get sample rate errors. Also some resource to download more MOH wav files.

Many thanks

[pete at privateconnect...]

I've spent the last 2-3 months on researching TTS and ASR for FS for a project.  Best TTS depends on what you consider important.  Also, how do you plan on using it.


Here's some of the TTS engines I've run across with some pros/cons:


Festivate Lite (flite)
Pros:
- Free (comes with FS)
- simple to use
- 16K voice sounds decent
- Completely customizable
Cons:
- 8K voice sounds horrible over cell phone


NeoSpeech (VoiceWare) (around $300/port for 1 voice + $75 each addl voice)
Pros:
- My selection for best soundig voices
- Recently select by Stephen Hawkings for his voice (geek points!)
- Lots of Languages supported
- Free trial available
Cons:
- Custom C-Based API (FS interface coming soon)
- Large file size (Engine + SDK + 1 Voice = 900MB)
- Support is lacking (Company beed in Korean, time zone issues, etc)


Nuance ($500/port for 1 voice)
Pros:
- Wide Variety of Products
- Support MRCP
- Supports ASR as well (add'l fees)
- Excellent support
- Free trial
- Decent sounding voices at 8K and 16K
- Wide range of tuning parameters
Cons:
- Pricey
- Limited voice selection
- Limited support for 64-bit linux


AT&T (NaturalVoice) (no pricing info available)
Pros:
- Big company (solid in marketplace)
- Good suppport (user and developer)
- ASP model means no software to maintain
Cons:
- ASP model incurs delay
- Voices sound too digitized
- Limited support for 64-bit linux



Loquendo ($500/port for 1 voice + 15% addl voice)
Pros:
- Good sounding voices (almost as good as NeoSpeech)
- Wide variety of languages
- Excellent support
- Has free 30 day trial
- Supports MRCP
- Support ASR and Voice Recognition as well. (add'l fees)
- Small footprint (< 150MB)
Cons:
- Pricey
- Complicated install process
- Limited management/tuning capabilities


In the end, it was down to NeoSpeech or Loquendo for our application.  We are currently running tests with NeoSpeech and assuming all goes well, we will select them.  Though don't let that color your opinion too much after several "focus groups" we discovered the most important element in the equation is does your customer/boss like the sound of the voices, and that is a completely subjective decision.




-pete

[dave at 3c.co.uk]

Just to add one:

Cepstral - quite a lot cheaper than the non-free ones Pete mentioned;
voice quality quite adequate for what we want it for, which is short
prompts and development. Not tried it for reading e-mail or the like.
FS' mod_cepstral isn't wholly compatible with their 5.x release (unless
someone's fixed it), but that's easily worked around.

--Dave

[Prometheus001 at gmx.net]

Thanks for this overwiev.

One question: How does this compare to Cepstral TTS?

Best regards
Peter

pete@privateconnect.com schrieb:

[saeedahmad1981 at gmai...]

Thanks guys for a detailed reply specially pete.


On Tue, May 19, 2009 at 7:31 PM, Peter P GMX <Prometheus001@gmx.net (Prometheus001@gmx.net)> wrote:

[mszlazak at aol.com]

Neospeech does have the best voices.

I looked at Neospeech months ago and talked to a rep.
Then he quoted me a price per port of over $1000.00. Looks like they really have done some big price adjustments.



-----Original Message-----
From: Saeed Ahmad <saeedahmad1981@gmail.com>
To: freeswitch-users@lists.freeswitch.org
Sent: Fri, 22 May 2009 2:01 am
Subject: Re: [Freeswitch-users] text to speech IVRs and MOH

Thanks guys for a detailed reply specially pete.


On Tue, May 19, 2009 at 7:31 PM, Peter P GMX <Prometheus001@gmx.net (Prometheus001@gmx.net)> wrote:

[pete at privateconnect...]

Just to followup on Cepstral. I used Cepstral for an Asterisk project a while back. So my information may be dated. I had not considered the for this project based on quality of voices.

Pros:
- Cheaper than other solutions ($50/port, $30/voice)
- Well documented
- Good Support
- Supports MRCP
- Has ASP Model
Cons:
- Voices sounded a bit too digital
- Limited Languages available
- Somewhat difficult install
- Lacking a wide range of tuning options