Ich Sage Was Ich Sagen Kann (ISWISK): Voice Composition by Real-time Automatic Speech Recognition

Abstract

Ich Sage Was Ich Sagen Kann is a voice composition experiment, where multiple self-made voice bots have a dialogue with each other by using Keyword Spot- ting (KWS) in the Automatic Speech Recognition (ASR) domain. The dialogue is not considered as a conversation between different personalities, but it reflects on a process of uttering thoughts from my personal experience. I make speech recognition modules by using a Convolutional Neural Network (CNN) model in- stalled on eight microcontrollers, which give utterances when certain keywords are detected. The composition consists of four parts, from polite one-on-one conversation between myself and one bot to the impolite conversation between 8 bots, which reflects the chaotic state of mind in the thinking process. This practice aims not only at overcoming problems in communication and learning of a foreign language at a personal level, but also at the possibility of communica- tion with human language between machines.