MBRØLOID is a project to replace non-free voice synthesizers such as MBROLA and VOCALOID. It's based on a MultiBand-Resynthesis Vocoder and uses human voice recordings, so it can produce a more natural sound. By design it supports many different languages, but the first voices to be recorded will only support Japanese and English. It does not include a phonetic preprocessor, but it can be used by any frontend that support the SAMPA phonetic alphabet.

The core component is a drop-in replacement for the mbrola executable that can be used with eSpeak. Voicebanks to be created for MBRØLOID will use the same phoneset as existing MBROLA voices so that eSpeak can remain unmodified. After recording and labeling the voicebank will be compacted to a single file. The source tree of each voicebank will be released so that users can fix bugs in the voicebank and base new voicebanks on existing ones.

For singing synthesis sustained vowels need to be recorded and the syntheizer needs to be modified to use a sustained vowel for long notes. It makes also sense to record at different pitches and tempos. To test the ability to synthesize voice in real-time a MIDI-keyboard controlled synthesizer will have to be implemented. Finally we will write a plugin for QTau, a clone of the VOCALOID editor so that users can start producing songs. It is also possible to use existing voices of the Festival Speech Synthesis System by adding a singers formant.


This website is available for offline use from a git repository. RSS feed for this site