From data
To meaning


Advanced Speech technology

Text-O-Phone (ToP) is the complete, flexible, easy to integrate multilingual front-end solution for text-to-speech systems developed by CELI. It covers the complete processing pipeline form standard text to phonetic annotation – including stress and prosodic information.

Why choose Text-O-Phone:

  • Proprietary technology
  • Based on highly qualified language skills
  • Easy integration
  • Higher quality / lower TCO compared to open source alternatives


Text-O-Phone is a suite of multilingual, Natural Language Processing enabled modules, each one responsible for a different processing steps:

  • SSML input processing
  • User-provided pronunciation and normalization rules
  • Tokenization Expansion of non-standard words
  • Morphological analysis and POS annotation
  • Disambiguation
  • Chunking and phrasing
  • Phonetic transcription (grapheme-to-phoneme and phoneme-to-phoneme)
  • Annotation of suprasegmental features (prosody, intonation and pauses))

Modules are designed as language-independent software components. They can be specialized for a particular language by using language-specific resources and set-up.

Covered languages

To date, Text-O-Phone is provided for the following languages:

  • Italian
  • Brazilian Portuguese
  • English (American, British, Canadian)
  • Danish
  • Dutch
  • Finnish
  • French and French-Canadian
  • German
  • Greek
  • Hungarian
  • Indonesian
  • Mexican-Spanish
  • Norwegian
  • Polish
  • Romanian
  • Russian
  • Slovak
  • Swedish

We are currently extending the list of covered languages. Moreover, we can develop new language modules on request, along with the development of specialized languages for particular domains or areas.

Language resources

Linguistic analysis is supported by the following resources:

  • Phonetic dictionaries
  • Text corpora covering a variety of domains
  • Morphological databases
  • NSW collections (set of rules)
  • Manually verified test suites

Technical Specifications


  • Supports SSML say-as, sub and phoneme input parameters
  • IPA or X-SAMPA output format written in ANSI C to obtain a very high computational performance/memory footprint ratio
  • Source code and source-code generation tools available
  • Designed for mobile and embedded systems
  • Multi-platform (Windows and Linux) which can be integrated via ANSI C API
  • Accuracy up to 99%*, measured against a reference test suite annotated by two native speakers
  • Average speed: 75 thousand bytes per second, on a reference Intel(R) Xeon(R) CPU E3-1230 V2 @ 3.30GHz
  • Size: as low as 2 MB, compiled at 32 bit (varies depending on the language)
  • Memory usage: as low as 0.6 MB (varies depending on the language)

*Mexican and French Canadian libraries

Contact us for a demo

    I have read and agree to the Privacy Policy