A lady with a surprised face listens to a laptop

Natural? Expressive? Perfect!

domainTTS is an added-value speech service by e-rhetor that delivers a complete solution for the automated conversion of text to premium quality speech for applications with well-defined lexical and grammatical domains. domainTTS is based on pesto TTS text-to-speech synthesizer and adds custom-built natural and expressive domain voices that are built on demand. Indicative examples include:

  • Customer care (e.g. account balance, actions, …)
  • Weather forecast
  • Game results
  • Routes Arrivals/Departures
  • Gas stations/pharmacies/firms search
  • Announcement
  • Alarms

Comparison with other TTS

A typical TTS system rarely achieves the desired speech expressions. The following table compares the domainTTS with other commercial "natural voice" speech synthesis solutions.

FeatureOther TTSdomainTTS
Expressiveness No Yes
Questions No Yes
Unique/exclusive voice No Yes
Audio quality Very good Perfect
Unlimited lexicon Yes No
Time to deliver 0 from few hours
If an existing voice

Comparison with waveform concatenation

The most common alternative to domainTTS is the concatenation of small pieces of pre-recorded utterances. Such a technique is widely used, especially in IVR applications. But, things get annoying for customers contacting automated call centers, when listen to "broken", thus unnatural, speech prompts. The comparative advantages of domainTTS are:

  • the lack of pauses and gaps in between words
  • the maintaining of the prosody naturalness
  • the audio quality of the generated waveform
  • the flexibility of domainTTS to synthesize new prompts

Example: an IVR time-announcement application

The time is 25 past 8. (note: in Greek this is "Η ώρα είναι 8 και 25.")

Utternace pattern: Η ώρα είναι [1-2]?[0-9] (και [0-9][0-9]?)?