PerSynt

(very tentative project sketch)

The PerSynt project

Personal Synthesis -- Personlig Syntes

The PerSynt project aims to create personal synthesis, that is synthesis of anyone's voice.

People want more variation in text-to-speech.

People want to have personal voices, even their own voice.

Different voices may even be necessary, for example in storytelling, developing voices for animations, games etc.

This project aims to develop tools, knowledge, lexicons and other resources both to build personal voices and to understand the processes involved.

Labelling

  • Unnecessary/Irrelevant work that shouldn't be done manually:
    • locating files
    • loading files
    • saving (label) files
    • searching in the speech waveform
  • Relevant/Useful
    • waveform
    • spectrogram
    • what was said?

Possible outcomes

  • lexicon (pron, pros, pos)
  • indexing of a number of speakers in the swedia db (pron, pros, pos)
  • duration models for a number of speakers
  • intonation models for a number of speakers
  • lexical reduction models for a number of speakers
  • synthetic voices for a number of speakers
    • even though these may have limited public or commercial use because of the licencing of swedia voices, but knowledge, experience and tools that will facilitate the constructions of such voices in the future.
  • combined models ('alla äldre män i småland' etc)
  • comparisons (intonation, lex red)
  • mixed models (eg skåne-stockholm)

Research issues

  • synthesis from spontaneous speech
  • synthesis with personal characteristics
  • what is a 'voice'?