Abstract

Speech signals received over the public telephone network are increasingly being used in broadcasting, particularly for contributions from correspondents and commentators and in the "phone-in" type of programme. Though telephone - speech quality is normally adequate for communicating information it can compare very unfavourably with studio speech in the same broadcast programme. This Report considers the possibility of making broadcast telephone speech more acceptable by using speech synthesis to extend the bandwidth beyond the 300Hz to 3.4kHz limits commonly set by the telephone system. Tests have indicated that if the missing components below 300Hz and above 3.4 kHz could be synthesised ideally a marked improvement would be achieved. Further tests showed that a practical synthesis of low-frequency components based on pitch information extracted from simultaneously recorded wide-band speech gave a smaller but distinct improvement. However, low-frequency synthesis based on pitch information extracted from the available telephone signal was markedly less acceptable - often sounding rough, and failing to blend with the telephone signal. This result is attributed to multiple, missing, or irregularly-spaced pitch pulses. It is concluded, therefore, that telephone signal band-width extension would be worthwhile if it could be achieved satisfactorily, but that a very precise pitch extractor is required for the low-frequency speech synthesis process.