Proceedings Fonetik 2009 - Institutionen för lingvistik

More documents

Recommendations

Info

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm UniversityModified re-synthesis of initial voiceless plosives byconcatenation of speech from different speakersSofia StrömbergssonDepartment of Speech, Music and Hearing, School of Computer Science and Communication, KTH,StockholmAbstractThis paper describes a method of resynthesisingutterance-initial voiceless plosives,given an original utterance by onespeaker and a speech database of utterances bymany other speakers. The system removes aninitial voiceless plosive from an utterance andreplaces it with another voiceless plosive selectedfrom the speech database. (For example,if the original utterance was /tat/, the resynthesisedutterance could be /k+at/.) In themethod described, techniques used in generalconcatenative speech synthesis were applied inorder to find those segments in the speech databasethat would yield the smoothest concatenationwith the original segment. Results froma small listening test reveal that the concatenatedsamples are most often correctly identified,but that there is room for improvement onnaturalness. Some routes to improvement aresuggested.IntroductionIn normal as well as deviant phonological developmentin children, there is a close interactionbetween perception and production ofspeech. In order to change a deviant (non-adult)way of pronouncing a sound/syllable/word, thechild must realise that his/her current productionis somehow insufficient (Hewlett, 1992).There is evidence of a correlation between theamount of attention a child (or infant) pays tohis/her own speech production, and the phoneticcomplexity in his/her speech production(Locke & Pearson, 1992). As expressed bythese authors (p. 120): “the hearing of one’sown articulations clearly is important to theformation of a phonetic guidance system”.Children with phonological disorders producesystematically deviant speech, due to animmature or deviant cognitive organisation ofspeech sounds. Examples of such systematicdeviations might be stopping of fricatives, consonantcluster reductions and assimilations.Some of these children might well perceivephonological distinctions that they themselvesdo not produce, while others have problemsboth in perceiving and producing a phonologicaldistinction.Based on the above, it seems reasonable toassume that enhanced feedback of one’s ownspeech might be particularly valuable to a childwith phonological difficulties, in increasinghis/her awareness of his/her own speech production.Hearing a re-synthesised (“corrected”)version of his/her own deviant speech productionmight be a valuable assistance to the childto gain this awareness. In an effort in this direction,Shuster (1998) manipulated (“corrected”)children’s deviant productions of /r/, and thenlet the subjects judge the correctness andspeaker identity of speech samples played tothem (which could be either original/incorrector edited/corrected speech, spoken by themselvesor another speaker). The results fromthis study showed that the children had mostdifficulties judging their own incorrect utterancesaccurately, but also that they had difficultiesrecognizing the speaker as themselves intheir own “corrected” utterances. These resultsshow that exercises of this type might lead toimportant insights to the nature of the phonologicaldifficulties these children have, as wellas providing implications for clinical intervention.Applications of modified re-synthesisApart from the above mentioned study byShuster (1998), where the author used linearpredictive parameter modification/synthesis toedit (or “correct”) deviant productions of /r/, amore common application for modified resynthesisis to create stimuli for perceptual experiments.For example, specific speech soundsin a syllable have been transformed into intermediateand ambiguous forms between twoprototypical phonemes (Protopapas, 1998).These stimuli have then been used in experimentsof categorical perception. Others havemodulated the phonemic nature of specificsegments, while preserving the global intonation,syllabic rhythm and broad phonotactics ofnatural utterances, in order to study what198
Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm Universityacoustic cues (e.g. phonotactics, syllabicrhythm) are most salient in identifying languages(Ramus & Mehler, 1999). In these typesof applications, however, stimuli have been createdonce and there has been no need for realtimeprocessing.The computer-assisted language learningsystem VILLE (Wik, 2004) includes an exercisethat involves modified re-synthesis. Here,the segments in the speech produced by theuser are manipulated in terms of duration, i.e.stretched or shortened, immediately after recording.At the surface, this application sharesseveral traits with the application suggested inthis paper. However, more extensive manipulationis required to turn one phoneme into another,which is the goal for the system describedhere.PurposeThe purpose of this study was to find out if it isat all possible to remove the initial voicelessplosive from a recorded syllable, and replace itwith an “artificial” segment so that it soundsnatural. The “artificial” segment is artificial inthe sense that it was never produced by thespeaker, but constructed or retrieved fromsomewhere else. As voiceless plosives generatedby formant synthesizers are known to lackin naturalness (Carlson & Granström, 2005),retrieving the target segment from a speech databasewas considered a better option.MethodMaterialThe Swedish version of the Speecon corpus(Iskra et al, 2002) was used as a speech database,from which target phonemes were selected.This corpus contains data from 550adult speakers of both genders and of variousages. The speech in this corpus was simultaneouslyrecorded at 16 kHz/16 bit sampling frequencyby four different microphones, in differentenvironments. For this study, only therecordings made by a close headset microphone(Sennheiser ME104) were used. No restrictionswere placed on gender, age or recording environment.From this data, only utterances startingwith an initial voiceless plosive (/p/, /t/ or/k/) and a vowel were selected. This resulted ina speech database consisting of 12 857 utterances(see Table 1 for details). Henceforth, thisspeech database will be referred to as “the targetcorpus”.For the remainder part of the re-synthesis, asmall corpus of 12 utterances spoken by a femalespeaker was recorded with a Sennheiserm@b 40 microphone at 16 kHz/16 bit samplingfrequency. The recordings were made in a relativelyquiet office environment. Three utterances(/tat/, /kak/ and /pap/) were recorded fourtimes each. This corpus will be referred to as“the remainder corpus”.Table 1. Number of utterances in the target corpus.Nbr of utterancesUtterance-initial /pV/ 2 680Utterance-initial /tV/ 4 562Utterance-initial /kV/ 5 614Total 12 857Re-synthesisEach step in the re-synthesis process is describedin the following paragraphs.AlignmentFor aligning the corpora (the target corpus andthe remainder corpus), the NALIGN aligner(Sjölander, 2003) was used.Feature extractionFor the segments in the target corpus, featureswere extracted at the last frame before the middleof the vowel following the initial plosive.For the segments in the remainder corpus, featureswere extracted at the first frame after themiddle of the vowel following the initial plosive.The extracted features were the same asdescribed by Hunt & Black (1996), i.e.MFCCs, log power and F0. The Snack toolSPEATURES (Sjölander, 2009) was used toextract 13 MFCCs. F0 and log power were extractedusing the Snack tools PITCH andPOWER, respectively.Calculation of join costJoin costs between all possible speech segmentcombinations (i.e. all combinations of a targetsegment from the target corpus and a remaindersegment from the remainder corpus) were calculatedas the sum of1. the Euclidean distance (Taylor, 2008) in F02. the Euclidean distance in log power199
Page 1 and 2:
Department of LinguisticsProceeding
Page 3 and 4:
Proceedings, FONETIK 2009, Dept. of
Page 5 and 6:
Page 7 and 8:
Page 9 and 10:
Page 11 and 12:
Page 13 and 14:
Page 15 and 16:
Page 17 and 18:
Page 19 and 20:
Page 21 and 22:
Page 23 and 24:
Page 25 and 26:
Page 27 and 28:
Page 29 and 30:
Page 31 and 32:
Page 33 and 34:
Page 35 and 36:
Page 37 and 38:
Page 39 and 40:
Page 41 and 42:
Proceedings, FOETIK 2009, Dept. of
Page 43 and 44:
Page 45 and 46:
Page 47 and 48:
Page 49 and 50:
Page 51 and 52:
Page 53 and 54:
Page 55 and 56:
Page 57 and 58:
Page 59 and 60:
Page 61 and 62:
Page 63 and 64:
Page 65 and 66:
Page 67 and 68:
Page 69 and 70:
Page 71 and 72:
Page 73 and 74:
Page 75 and 76:
Page 77 and 78:
Page 79 and 80:
Page 81 and 82:
Page 83 and 84:
Page 85 and 86:
Page 87 and 88:
Page 89 and 90:
Page 91 and 92:
Page 93 and 94:
Page 95 and 96:
Page 97 and 98:
Page 99 and 100:
Page 101 and 102:
Page 103 and 104:
Page 105 and 106:
Page 107 and 108:
Page 109 and 110:
Page 111 and 112:
Page 113 and 114:
Page 115 and 116:
Page 117 and 118:
Page 119 and 120:
Page 121 and 122:
Page 123 and 124:
Page 125 and 126:
Page 127 and 128:
Page 129 and 130:
Page 131 and 132:
Page 133 and 134:
Page 135 and 136:
Page 137 and 138:
Page 139 and 140:
Page 141 and 142:
Page 143 and 144:
Page 145 and 146:
Page 147 and 148: Proceedings, FONETIK 2009, Dept. of
Page 197: Proceedings, FONETIK 2009, Dept. of
Page 203 and 204: Proceedings, FOETIK 2009, Dept. of
Page 205 and 206: Proceedings, FOETIK 2009, Dept. of
Page 227: Department of LinguisticsPhonetics
show all

Proceedings Fonetik 2009 - Institutionen för lingvistik

Create successful ePaper yourself

Delete template?

Save as template?