Hi Philip, When it comes to over all atmosphere you are right. A human voice is always superior, but I haven't been doing that with any of my games. With my original draft of STFC I used Neosspeach Kate, then Realspeak Karen, and in many of the betas of MOTA I was using Acapela Heather. Since all of these games were essentially using prerecorded clips of SAPI voices anyway I was essentially doing a lot of unnecessary work recording,editing, and then using the wav files of synth voices when I could have used them directly.
Now, my voice isn't very good, and do to some physical impairments it is impossible for me to speak clearly and I would never use it for a professional game. Unfortunately, that means I would have to hire out to someone else to do the voice clips and I don't really want to spend a great deal of money on a game like Mysteries of the Ancients just to get realistic human speech. Perhaps if the game sells well I could consider reinvesting some of that money in voice acting etc, but for a preliminary release SAPI is definitely the best way to go. As far as wrapper classes that's pretty much what I was doing from betas 1 to 22. I had a function called SpeakNumber() that would take a number and do all the loading and processing of number files which works pretty well. However, that required some extra work developing those wrapper classes in order to get the same functionality I have with SAPI right away. There is one other issue you didn't mention and that is cross-platform support. There really aren't any good cross-platform TTS solutions. ESpeak, Festival, etc are all pretty bad and using prerecorded speech clips is a great way to support Windows, Linux, Mac, and iOS, etc all with the same speech and from that perspective it is easier to deal with than a specific API like SAPI. Cheers! On 5/4/13, Philip Bennefall <[email protected]> wrote: > Hi Thomas, > > I just wanted to respond with a few spontaneous thoughts based on my own > experiences with sapi versus prerecorded audio. > > Mostly I agree with your points. But for my own part I find that using > synthetic speech, whether it be prerecorded or generated on the fly by a > sapi controlled engine, greatly detracts from the atmosphere. Imagine if you > > will, that you are in a shadowy temple with a faint sound of wind, distant > screams from the depths of a dungeon far below, haunting music with > reverberating strings, and an equally dramatic narrative performed by > Microsoft Mike. A bit of a contrast. > > To take a less extreme example, in my upcoming title the atmosphere is > generally quite dramatic and with a dark undertone most of the time. > Therefore, when I recorded my speech files I spoke in a very low voice even > > for such trivial things as numbers and status messages. I knew in what > context my voice was going to be played, and was able to adapt it to really > > blend in with the rest of the scene that I was trying to create. A speech > engine could never achieve that. So I think that in terms of atmosphere and > > dramatic effect, prerecorded speech is far superior than a tts engine. There > > are, of course, numerous cases where the advantages of tts output greatly > outweigh the pros of prerecorded material, such as when there is absolutely > > no way for you to know what content that may need to be spoken. But if, as > in the case of Mota, you are using prerecorded speech files that just > contain tts generated content anyway then these have absolutely no advantage > > so I would not disagree with your decision to use sapi if you aren't > interested in getting a voice talent to record your files. In short, I feel > > that prerecorded voice files are only advantageous if you actually have a > real voice for the game. > > As for ease of coding - that has never really bothered me. It is trivial to > > design wrapper classes that can take a number, or a filename etc using > overloaded methods, and rendering that with the available speech files > asynchronously in the same thread so that one doesn't have to care about > synchronization. So with a careful design you can get it down to a one liner > > just like you would with sapi. What I find annoying with sapi is that I can > > never be sure what the final output will actually sound like. Many of the > commercial voices have quite a bit of delay during their initial buffering > for each new phrase. When you speak the same phrase for the second time they > > are usually faster, but the fact that there is so much variation between > voices makes it difficult to judge in advance how the game will actually > perform if that makes sense. I tend to use sapi extensively for prototyping > > a game, but once I reach a stage where I start to consider a potential > release date my first priority is to rip out sapi and replace it with real > human speech. > > In summary, I think that if you want to get something out quickly and > without spending any money then sapi is definitely the right way to go. If > on the other hand you are interested in maintaining a great atmosphere and > you want to make sure the game sounds and performs the same everywhere, I'd > > recommend a voice actor. Best of luck, and sorry about the extremely lengthy > > ramble! > > Kind regards, > > Philip Bennefall --- Gamers mailing list __ [email protected] If you want to leave the list, send E-mail to [email protected]. You can make changes or update your subscription via the web, at http://audyssey.org/mailman/listinfo/gamers_audyssey.org. All messages are archived and can be searched and read at http://www.mail-archive.com/[email protected]. If you have any questions or concerns regarding the management of the list, please send E-mail to [email protected].
