Conversational speech systems for on-board car navigation and assistance.

摘要: This paper describes our latest efforts in building a speech recognizer for operating navigation system through instead of typed input. Compared to conventional recognition systems, where the input is usually restricted fixed set keywords and keyword phrases, complete spontaneous sentences are allowed as We will present interaction input, parsing necessary reactions requested queries. Our has been trained on German data adapted queries using MLLR. As not command word parser further process recognized utterance. show that within lab environment able handle arbitrary successfully. The performance measured error rate gives result 18%. also evaluated yields an 20%.