LING 15 Lecture Notes - Lecture 23: Speech Synthesis, Formant, Speech Recognition
Document Summary
(cid:272)o(cid:374)so(cid:374)a(cid:374)ts (cid:862)gau(cid:272)ho(cid:271)a(cid:272)k(cid:863) is [gautfo(cid:271)aek]; ge(cid:374)erati(cid:374)g pho(cid:374)e(cid:373)es from letters is tricky (though, tough, cough, though) Word pho(cid:374)e(cid:373)i(cid:272) stri(cid:374)g audio (cid:272)hu(cid:374)ks. (cid:863)e(cid:454)(cid:272)eptio(cid:374) (cid:272)he(cid:272)k(cid:863) list of irregular spelled for(cid:373)s & pho(cid:374)e(cid:373)i(cid:272) strings. Regular forms: generate phonemic strings according to rule. Conversion to audio portions -not segment-by-seg(cid:373)e(cid:374)t, i(cid:374)stead it(cid:859)s transition-by-tra(cid:374)sitio(cid:374) he+el+lo i(cid:374)stead of h+e+l+o. Synthesized or extracted from recordings; either way, program looks up what to produce. Speak-and-spell: talks out loud & tells you what words to type/spell. Sam (software automatic mouth): synthetic speech sounds robotic; predict pronunciation from spelling. Large numbers of files/subroutine but not as many as word-based. Synthesis reduces perceptibility; snippets from real speaker (e. g. siri) A(cid:455) (cid:862)gau(cid:272)ho(cid:863) to iri iri (cid:449)ill hear (cid:858)[ga] [ou] [(cid:272)ho] Fluctuations in ap reinterpreted as frequency & loudness. Parser: some program or device that detects symbolic categories (phonemes or words) from audio inputs. Like spee(cid:272)h ge(cid:374)eratio(cid:374), aso (cid:374)eeds a (cid:862)look-up(cid:863) step parser takes input signal & chooses an interpretation.
