Question

I seek some insight in creating an application that converts text to speech in ASP.NET. From my initial research, it appears that:

1. MS SAPI requires the client to download an ActiveX component and can support large amounts of text to be converted. Our clients are not willing to install any components on their systems, so this approach may or may not fly.

2. I do understand with .NET 3.0, we have the System.Speech.Synthesis namespace. Does the conversion take place on the server? If so, how would I serve it to the client?

Our requirements are ability to convert large amount of text, should be scalable and reliable. Which technology is “production ready” capable of serving a large number of requests in a short time interval.

Any thoughts are appreciated.

Answer 1

By default, ASP.Net applications don’t run with sufficient permissions to access Speech Synthesis, and attempting to run Larsenal’s code will fail with a security error.

I was able to get around this in an app by having a separate WCF service running on the server, as a regular Windows Service. The ASP.Net application then communicated with that service. That service just wrapped Larsenal’s code, returning an array of bytes, given a string of text.

Also, one megabyte of text? That’s a good-sized novel.

Edit, 11-12-09, answering some comments:

System.Speech can either return an array of bytes, or save to a wav file, which you can then feed to a media player embedded on the user’s page. When I built my talking web page, it worked like this:

1) Page.aspx includes an ’embed’ tag that puts a Windows Media Player on the page. The source is “PlayText.aspx?Textid=whatever”.
2) PlayText.aspx loads the appropriate text, and communicates (via WCF) to the speechreader service, handing it the text to read.
3) The Speechreader service creates a MemoryStream and calls SpeechSynthesiser.SetOutputToWaveStream, and then returns the stream as a single array of bytes. This array is Response.Write()-ed to the client.

Here’s the meat of the SpeechReader service:

    byte[] ITextReader.SpeakText(string text)
    {
        using (SpeechSynthesizer s = new SpeechSynthesizer())
        {
            using (MemoryStream ms = new MemoryStream())
            {
                s.SetOutputToWaveStream(ms);
                s.Speak(text);
                return ms.GetBuffer();
            }
        }
    }

I’m pretty sure that on the back end, this returns an enormous XML array-of-bytes, and is horribly inefficient. I just did it as a proof of concept, and so didn’t research that. If you intend to use this in production, make sure that it’s not internally returning something like this:

<byte>23</byte>
<byte>42</byte>
<byte>117</byte>
...