Speech Portal

Primetalk's Speech Portal is a platform for constructing sophisticated spoken dialog systems. It uses SynapseGrid as an integration platform and contains all the necessary components for a dialog system to speak with a human being:

1. Pipeline:
1. Speech signal processing.
2. Recognizer.
3. Parser.
4. Dialog manager.
5. Templater.
6. Synthesizer
2. Knowledge representation
1. A priori
1. Grammar expressions for input and output language.
2. Owl/rdf ontology import.
2. A posteriori
1. Strictly-typed frames.
2. Untyped frames.
3. Partial (at runtime) and uncertain
1. Fuzzy sets.
2. Incomplete frames.
3. Probability distributions.
3. Basic dialog logic building blocks
1. User input comprehension.
2. Context representation.
3. Output utterance proposals.
4. API for custom dialog logic
1. Extended Petry nets.
2. Composable functional speaking strategy.
5. API for real-time custom reactions.
1. Selected set of contacts (hooks) for getting all interesting real-time events.
2. Carefully designed replaceable parts of pipeline.

For recognizer part we support:

1. sphinx4 + our acoustic model or any voxforge models.
2. google speech api (unofficial).

For speech synthesis:

1. Festival + voice.
2. OpenMary + voice.
3. RhVoice + voice.
4. SpeechPro + voice.