Context, always context. A following vowel or voiced consonant tends to voice the , "speakerz-ov-English", an unvoiced consonant devoices it, "speakerss-seem-many". There are likely regional differences as well, but again, always depending on the surrounding context.