Speech and language processing

ARI can recognize speech, handle dialogues, and synthetise voice in several languages. It is fully compliant with the ROS4HRI REP-155 ROS standard.

The entire speech pipeline runs on-board; no cloud-based services are used (and consequently, no Internet connection is required).

The following figure provides complete picture of the speech processing pipeline on ARI.

The main building blocks are:

• ARI’s reSpeaker microphone;

• the Vosk speech recognizer. Learn more about speech recognition on ARI;

• a node monitoring incoming speech for wake-up/sleep keywords;

• the RASA dialogue manager. Learn more about dialogue management;

• the acapela speech synthesis library. Learn more about speech synthesis.

General documentation

• Dialogue management
• Wake-up word detector
• 🚧 How to design and trigger actions from the chatbot

How-to

• 🚧 How-to: Text-to-speech on ARI
• How-to: Automatic Speech Recognition (ASR) on ARI

Tutorials

• Tutorial: create or update a chatbot for ARI
• 🚧 Tutorial: Programming an interactive a story
• 🚧 Tutorial: Using the knowledge base with the chatbot

FAQ

• How to change the ‘wake-up’ (or ‘sleep’) keywords?

References

• Speech and language processing capabilities
• ASR, TTS and dialogue management APIs
• Default chit-chat/smalltalk capabilities of ARI