A telephone speech recognition system is provided which has a high level of speech recognition without affecting from various conditions of a telephone line. The system comprises speech analyzers (4) and (5), and reference speech model storages (7) to (9) corresponding to line connection data. A telephone line interface (1) having a line connection data acquisition function analyzes a call received from the telephone line for identifying the country, route, and other information of the call and transmits those line connection data to a line connection data processor (2). The line connection data processor (2) selects one of the acoustic analyzers (4) and (5) in response to the line connection data from the interface (1) and also one of the speech model storages (7) to (9). A speech pattern matcher (11) compares an acoustic vector train output of the selected acoustic analyzer with the speech models given from the selected reference speech model storage for speech recognition.