Fundamental frequency measures in the comparison of habitual and disguised voicesABSTRACT
Purpose This study aimed to investigate the stability of measures of fundamental frequency of voice by comparing samples of habitual and disguised speech.
Methods This observational, analytical, cross-sectional study analyzed a voice bank of 40 people of both sexes (20 men and 20 women), aged 19 to 58 years. They read a passage in both habitual and disguised speech. The types of disguises were analyzed by three expert voice judges. The study used the PRAAT® acoustic analysis software to extract the mean f0, median f0, minimum f0, maximum f0, and base f0 (fb) from the complete reading in both situations, using pre-determined procedures. The data were tabulated and statistically analyzed.
Results The fb, maximum f0, and minimum f0 values were similar in habitual and disguised speech among speakers of both sexes, whereas the mean f0 and median f0 differed between the two situations.
Conclusion In general, the measures of fb, minimum f0, and maximum f0 were more stable in disguised speech, which may suggest future usefulness in intraindividual comparisons. The speakers’ various types of adjustments to disguise their voices did not significantly alter the measures of fb, maximum f0, and minimum f0 in the comparison between habitual and disguised speech.
Keywords:
Voice; Voice Quality; Speech Acoustics; Speaker Recognition Interface; Acoustics
RESUMO
Objetivo Este estudo teve o objetivo de investigar a estabilidade das medidas de frequência fundamental da voz, comparando amostras de fala habitual e de fala em situação de disfarce.
Método Trata-se de estudo observacional, analítico e transversal. Foi analisado um banco de vozes de 40 pessoas de ambos os sexos (20 homens e 20 mulheres) e faixa etária entre 19 e 58 anos, correspondentes à leitura de trecho em fala habitual e leitura do mesmo trecho em fala em situação de disfarce. Os tipos de disfarces foram analisados por três juízas especialistas em voz. Além disso, foi utilizado software de análise acústica PRAAT® para a extração das medidas acústicas de f0 média, f0 mediana, f0 mínima, f0 máxima e f0 de base (fb) da leitura na íntegra, em ambas as situações, por meio de procedimentos pré-determinados. Os dados foram tabulados e analisados estatisticamente.
Resultados Os valores das medidas de fb, f0 máxima e f0 mínima foram semelhantes nas situações de fala habitual e disfarce, para locutores de ambos os sexos. Já os valores das medidas de f0 média e f0 mediana tiveram diferença nas duas situações.
Conclusão De modo geral, as medidas de fb, f0 mínima e f0 máxima mostraram maior estabilidade frente à situação de disfarce, o que pode sugerir utilidade futura em comparações intraindividuais. Os variados tipos de ajustes feitos pelos locutores para a realização dos disfarces não alteraram de modo significativo as medidas de fb, f0 máxima e f0 mínima quando as situações de fala habitual e disfarçada foram comparadas.
Descritores:
Voz; Qualidade da Voz; Acústica da Fala; Interface para o Reconhecimento da Fala; Acústica
INTRODUCTION
Forensic speaker comparison (FSC) is an examination indicated in cases where it is necessary to determine the authorship of speeches possibly linked to a crime that is recorded on media(1), comparing stored speeches with the suspect’s speech records. This comparison between materials makes it possible to arrive at a result that confirms or refutes the hypothesis of authorship. In most cases, the traces used in the forensic examination come from audio recordings originating from fixed or mobile telephone interceptions, environmental recordings made and authorized by the courts, audios recorded by one of the interlocutors in telephone communication, and from messaging applications (for example, WhatsApp)(1).
Acoustic analysis is fundamental in speaker comparison and a premise of international consensus, as it considers quantitative aspects and the possible visual association. The former includes objective values of various parameters, which can be analyzed both within and between speakers(2,3), whereas visual association is made possible through broadband and narrowband spectrograms and lends credibility to the analysis(4-8). Acoustic analysis is an important benchmark in comparing speakers, as it allows for greater understanding and comparison of the vocal signal, associations with auditory-perceptual evaluation, and visual documentation of voice and speech parameters(9).
Fundamental frequency of the voice (f0) is one of the acoustic measures frequently used in FSC(10-14), corresponding to the number of complete cycles of vocal fold opening and closing(9). Research carried out at the University of York(9) indicated that f0 is the most used acoustic measure in FSC and that 94% reported using arithmetic mean measures, 72% standard deviation, 41% median, 34% mode, 25% base f0 (fb), and 6% reported analyzing the range of f0 values. The widespread use of f0 in this examination is explained by its qualities and for being measurable and available within a speech sample(10).
F0, the number of vibrational cycles of the vocal folds per second, has a physiological characteristic related to vocal fold length, which can produce population averages(11). This frequency is the best known and most used parameter when it comes to voice recognition and speaker comparison(12-16). Moreover, the most frequent forms of disguise directly affect f0 (11), which indicates frequency variations in the voice, such as high and low pitch(15).
The speech style, vocal effort, and emotional state can affect f0, generating intra-speaker variation and decreasing its discriminative power(1). A study reported that only 25% of those consulted use the long-term measure of f0 (LT f0) – i.e., fb values(9). However, recent research has indicated that this measure seems to be more stable than other f0 parameters(16,17).
Fb, proposed by Swedish phoneticians Traunmüller and Eriksson(18), is a specific statistical estimator of location for f0 samples. Fb is the frequency of vibration of the vocal folds in a relaxed position, the frequency to which the vocal folds naturally return after a prosodic excursion. According to modulation theory, there must be a base value of the fundamental frequency, whose frequency represents the individual articulation of the speaker. Hence, it would be the best predictor of the individual's intrinsic f0 value, corresponding to the personal frequency of the vocal folds, the neutral f0, characteristic of the speaker(19).
The results of a recent study demonstrated that fb is the best parameter to be used in FSC f0 analyses(1). According to the authors, the statistical measure of the f0 baseline value is less affected by the speech style and content or the channel used in the recording and requires less audio to obtain a stable measure than commonly used f0 measures, including the arithmetic mean and standard deviation.
Studies are scarce in the literature on the analysis of voice disguise in forensic phonetics(11,15,20,21). Nevertheless, they can be very useful for understanding evidence-based speaker comparison.
Considering that speaker comparison often encounters disguise(22-24), it is necessary to understand how such comparisons are carried out. Giller(25) defines disguise as “[…] the deliberate action of a speaker who alters their voice, speech, or language to conceal their identity”. From this definition, we can understand that disguise is an imitation of a speech style or accent to hinder comparisons of that subject’s voice. In forensics, disguise appears most often in cases of kidnapping, extortion, harassment, prank calls, drug trafficking operations, homicides, corruption, and fraud. Voice disguise in a possible crime is one of the factors that most hinders forensic analysis(25).
Disguise can occur at the moment of the “questioned recording,” which is the term used in forensics for compared recordings. It can happen when the act occurs or when the subject knows that they are being recorded and will be subjected to comparison examination, this being a moment used for comparison between records(4).
For forensic work, it is of paramount importance to seek a parameter that is little influenced by environmental conditions and disguise, since criminals seek strategies and resources that can camouflage their true identity in the face of future evidence.
A key point regarding voice disguise is that it is not self-exhausted – i.e., it can provide important information about the acoustic parameters and their behavior in disguise. Thus, when the speaker disguises their voice, it can be compared with their habitual voice(26). Therefore, the subject’s habitual voice must be known in order to analyze their disguised voice(11).
Given the growing demand for specialists in forensics, speech-language-hearing pathology uses knowledge applied to human communication in forensic science to deal with evidence left by a crime or an alleged crime. Forensic comparative analyses of the voice are among the various possible contributions of speech-language-hearing pathology to the field(4).
This research is relevant in that it contributes to greater clarification and elucidation on the possibility of using acoustic measures of disguised voices in speech, since the literature still has gaps on the subject. It is estimated that by highlighting the best measures of speech for the comparison of disguised voices, it will be possible to use them more assertively, as they are easily extracted in any environment(11,27,28).
Hence, this study aimed to investigate the stability of f0 measures of the voice, comparing samples of habitual and disguised speech.
METHODS
This is an observational, analytical, cross-sectional study. The analysis sample was obtained from a voice bank collected for a previous study. The research was approved by the Research Ethics Committee of the Universidade Estadual do Centro-Oeste, UNICENTRO, Paraná, Brazil, under number 4.752.583.
Sample
A database of speech samples from 40 people was analyzed, subdivided according to sex and age as follows: 20 women, aged 19 to 55 years (average age: 32 years and 3 months), and 20 men, aged 19 to 55 years (average age: 33 years and 4 months). The group consisted of university students, university professors, and experts, with no professional announcers.
The voice bank samples were collected in a laboratory with acoustic treatment on a Pentium Dual Core 5.300 2.60 GHz computer, 1.99 GB RAM, XP 2002 Service Pack 3 processor, M-Audio Fast Track Pro 4x4 external sound card, and AKG C 3000 B microphone.
The audio recording was collected in wave format, mono channel, sampling frequency of 44.1 kHz, and 16 bits. The duration of the samples varied according to each speaker’s reading/speaking speed, ranging from 0.40 seconds to 2 minutes per audio.
The recordings were made using the Audacity® program. The voices were collected inside an acoustic booth, where the microphone was located. For the recording, the speakers were instructed neither to vary their body position to avoid distancing from the microphone nor to incline their neck to avoid modifications in laryngeal movement.
The speech text collected and used in this research simulated a distress call during a phone call and was read by participants four times: twice with their usual voice and twice with a disguised voice. For disguised reading, they were instructed to try not to be recognized, using any resources they deemed necessary. Thus, speakers were free to choose a disguise they considered effective, as long as they did not use resources other than their articulatory organs (e.g., using their hands to occlude the nostrils or any other type of interference).
Habitual speech was collected twice, so that the subject could become familiar with the text. This research used the second speech recording, considered the habitual speech sample, and the first disguised recording because speakers generally used similar resources in both, not considering whether they could maintain the disguise.
Previous studies had already used the voice bank. Researchers in literature/linguistics and forensics developed the following text (consisting of 69 words in Portuguese) during a master's thesis: "Hello, I want to speak with Dona Teca. Dona Teca, this is the devil speaking. We have your husband's duck in the den, and we're going to kill him, chop him up, and throw him in a Coke bottle. Do you want to save him? Then don't make me angry. I want a million. Put everything in a package near the peccary's cask and run away. Don't call the police or I'll poke you and stab your head" (in Portuguese: "Alô, quero falar com a Dona Teca. Dona Teca, aqui fala é o capeta. Estamos com o pato do teu marido na toca e vamos matar ele, picar e tacar dentro de uma garrafa de Coca. Quer salvar ele? Então não me deixa puto. Quero um milhão. Bota tudo num pacote perto da pipa do cateto e se pica. Não chama a polícia senão te cutuco e espeto teu coco").
Its use in disguise considered various phonemes, especially intervocalic voiceless plosives [p, t, k], the onset of stressed syllables, the phonological context preceding the vowel [a], and the phonological context following [p]: [e, i, u]; following [t]: [a, e, ɛ, ɔ], and following [k]: [a, ɔ].
Disguise analysis
Three judges, experts in voice, with more than 15 years of experience in auditory-perceptual evaluation, including protocols involving the analysis of vocal tract adjustments, such as the John Laver Vocal-Profile Analysis Protocol (1980), were invited to analyze the types of disguises. The analyses proceeded as follows: initially, two of the three judges analyzed the samples of each subject, in pairs, individually, considering the subject's habitual voice as a baseline. Thus, the judges had to point out which resources/adjustments the speaker used to disguise their voice in comparison with their habitual voice.
The researchers developed a questionnaire containing 21 possible types of adjustments, based on common adaptations made by people in disguise speech, generally related to modifications in voice quality, frequency, intensity, articulation, resonance, and modulation. The judges were to mark an “x” when the speaker used the characteristic to disguise their voice.
The responses that differed between the two judges were sent to the third judge for tie-breaking and confirmation of whether the speaker used that adjustment. Moreover, 20% of the samples were repeated for intrarater agreement analysis. The kappa value obtained was greater than 0.8 for all of them, indicating substantial intrarater agreement.
The PRAAT® acoustic analysis software was used to extract f0 data, adopting the following steps for all samples from each participant (habitual and disguised): Importing the complete reading excerpt; analyzing the file's periodicity in relation to the pitch (f0), with adjustment of the parameters according to the speaker's sex; generating a specific file for f0 analysis; visually analyzing the f0 points of the generated file to eliminate spurious points and artifacts; extracting f0 measures from the corrected file: mean (sum of the values obtained in the set throughout the entire section, divided by the number of values summed), median (value that separates the larger half from the smaller half of a sample), fb (statistical measure that defines the frequency to which the vocal folds naturally return after a prosodic excursion)(18,19), minimum (lowest f0 value observed throughout a section) and maximum (highest f0 value observed throughout a section); and creating a distribution graph from the pitch file, verifying the f0 distribution.
Thus, the extraction followed these steps: 1. Adjusting the pitch analysis limits according to sex (75 to 300 Hz in males and 100 to 500 Hz in females); 2. Verifying the behavior of the pitch curve generated with the pulses considered automatically; 3. Not considering the production times of the segments of each individual, given that the calculations of f0 and fb are measured in Hz; 4. Creating pitch object; 5. Visually comparing the oscillogram, pulses, and pitch curve of the sound object with the points (samples) of the pitch object, excluding samples that did not correspond to a voice signal; 6. Obtaining the frequency values of the corrected pitch object; 7. Creating the distribution graph from the pitch object matrix.
The descriptive analysis of the results of the quantitative variables was performed using relative and absolute frequency. All numerical values obtained in the procedures described above were tabulated in a spreadsheet.
The mean, median, minimum, and maximum f0 values and fb values of habitual voices were compared with the mean, median, minimum, and maximum f0 values and fb values of disguised voices using statistical tests for dependent samples. Thus, the habitual speech sample was considered the control sample in relation to the disguised speech sample.
Statistical analysis
The study used Student's dependent t-test and the Wilcoxon test. The dependent sample tests were also used to compare the mean, median, minimum, and maximum fb during habitual and disguised speech, for men and women. Student's dependent t-test or Wilcoxon test was used, depending on the data distribution (normal or abnormal), which was analyzed with the Shapiro-Wilk normality test. In the case of statistically significant differences, the effect size was calculated using Cohen's d (variables with differences had a normal distribution). The interpretation of the effect size was as follows: 0.20 small effect, 0.5 medium effect, and 0.8 large effect.
All analyses used a 5% significance level, or 0.05.
RESULTS
Table 1 presents the characterization of the sample through descriptive analysis of the speakers’ vocal adjustments when disguising their voices in relation to their usual voices (Table 1).
Strained voice and greater loudness were the most used (n = 21; 52.50%), followed by lower pitch (n = 18; 45%), phoneme distortions/omissions/substitutions (n = 16; 40%), higher pitch, and pharyngeal constriction (n = 14; 35% in both). Although there were several higher relative frequencies in descriptive terms between men and women, the equality of proportions test, performed complementarily, indicated p > 0.05 in 20 of the 21 cross-matches between the sexes, with a difference only for pharyngeal constriction, which men used (n = 10; 50%) more than women (n = 4;20%) – p = 0.047.
The comparison between f0 results in disguised and habitual voices showed no differences in fb, minimum f0, or maximum f0. The same was not observed for the mean f0 (habitual 198.85 and disguised 216.62 – p = 0.021) or median f0 (habitual 195.31 and disguised 206.1 – p = 0.032), which were different between emissions, with higher values in disguised speech (Table 2). The additional calculation of Cohen's d indicated a small to medium effect for the mean f0 and median f0.
Comparison between the results of the fundamental frequency of the voice (f0) during habitual and disguised speech in the total group (n = 40)
The comparison between f0 results in disguised and habitual voices in females found no differences in any of the measures analyzed (Table 3).
Comparison between the results of the fundamental frequency of the voice (f0) during habitual and disguised speech among females (n = 20)
Finally, Table 4 shows the comparison between f0 results in habitual and disguised speech in males. This analysis found differences between the mean f0 (habitual 149.22 and disguised 182.75 – p = 0.09) and median f0 (habitual 146.27 and disguised 176.35 – p = 0.018). The table shows medium to large effect sizes of the mean f0 and median f0.
Comparison between the results of the fundamental frequency of the voice (f0) during habitual and disguised speech in males (n = 20)
DISCUSSION
Pitch changes during speech, going lower and higher, as it is the element responsible for intonation (frequency variation from the point of view of psychoacoustic perception)(15). Speakers vary their pitch because of the various tones used to express themselves. Hence, the f0 acoustic analysis allows the plotting of graphs called pitch contour curves to find instantaneous f0 values as a function of time(29).
The results of this study reinforce a previous study(15) regarding the common pitch variations in disguise speech, in that the lower pitch was the second most used resource, appearing in 45% of the disguises of the general, female, and male populations. The higher pitch also appeared among the speakers’ preferred disguises, with 32.5% of the general population. Pitch variations occurred in 7.5% of the disguise cases in this study in the general group, appearing only in the female population, in 15% of the disguises.
The authors of a recent study(30) analyzed the effects of disguise on the f0 and whether these affected the acoustic analyses. They observed that one of the most used types of adjustments was tract constriction, strongly associated with disguise f0 values in the control samples of men (p = 0.105) and women (p = 0.171). Thus, it can be understood that even with tract constriction, it is possible to perform acoustic analyses and ascertain whether the same subject is speaking.
Another type of disguise present in the study is the participant’s simulation of anger(30). The variations that anger can bring to disguise speech also appeared in this study, translating as greater loudness, which was the resource most used by the total group. All the resources regarding the previous study(30) were strongly associated with the disguise f0 values of the control of men and women samples – females had greater variations in disguise f0 values than males, a result that is divergent from the present study, in which women made smaller changes in f0 than men.
The opinion of experts has been increasingly requested and included in legal processes to answer whether a given voice sample belongs to the same speaker(3). This sample comparison and its results correspond to what is called forensic identification of the speaker(4). This study verified that some acoustic measures are more stable in the comparison between habitual and disguise speech samples of the same individual, which could be confirmed by the lack of statistically significant differences. This is evidenced by the similar fb, maximum f0, and minimum f0 results in both samples despite the speakers’ various vocal tract adjustments. These results indicate that such measures are not susceptible to change caused by vocal tract modifications, which may contribute to the expert report, confirming or refuting the hypothesis that it is the same speaker in the possible crime in question (Table 2).
These results indicate that fb, maximum f0, and minimum f0 change less than the other frequencies. This draws a parallel with another study(11), whose statistical analyses demonstrated differences in intra-speaker mean and median f0 when different types of disguises were simulated by varying speech style, emotions, vocal effort, and recording quality. Study authors(16) report that the results of mean f0, median f0, and standard deviation were affected by outliers – a term used in statistics to refer to atypical findings that deviate from normal, possibly harming the interpretation of the results of the samples in question. As for the present study, such outliers were removed to minimize this bias.
This study obtained similar fb, maximum f0, and minimum f0 results in habitual and disguised speech, despite the varied resources used by speakers. Fb is the frequency of vocal cord vibration in a relaxed position, the frequency to which the vocal folds naturally return after a prosodic excursion(19). Thus, the hypothesis is that it is less susceptible to diverse speech variations from a prosodic standpoint and from articulatory organ modifications, which can contribute to the expert report when comparing speakers in crime analysis.
To date, forensic f0 analysis has focused on its mean duration and standard deviation(15). The analysis of the present study results suggests that the mean and median duration may be less stable measures for disguised speech and sample comparison, as they present statistically significant changes. External factors, such as susceptible emotions, may greatly influence the mean f0 (19). Perhaps this could explain why the mean and median f0 differed between habitual and disguised speech. It is important to mention that the overall group had small to medium effect sizes in the comparison of the mean and median f0 in habitual and disguised speech – i.e., they are visible effects that may be relevant in clinical and/or forensic contexts. However, these same measures had medium to large effect sizes in males, suggesting that the differences found have significant practical relevance.
Several factors, didactically divided into technical, physiological, and psychological factors, influence and affect f0. Physiological factors are biological causes, such as age and smoking, while psychological factors are linked to emotions and environmental conditions at the time of emission. Technical factors, in turn, refer to the sample recording conditions(31). Nevertheless, f0 is a phonetic parameter with good results, being, therefore, considered reliable for forensic analysis(10). Stability in speaker comparison parameters is a key resource in forensic expertise; hence, f0 seems to meet a wide range of these criteria(31). Having f0 stability in mind, this study approached other f0 measures, strengthening the idea mentioned earlier(16,17) that fb is more stable than other f0 parameters, such as the already mentioned mean f0 and median f0.
This knowledge demonstrates that the expert comparison of disguised speech can use fb along with the maximum and minimum f0, which have yielded reliable results in forensic analysis in cases of crimes or suspected crimes.
Regarding limitations, this study did not evaluate the natural f0 variability of the same speaker in undisguised speech. This could contribute to assessing how much the effects observed in this study were attributable to vocal disguise or fluctuations in vocal emission. Future studies are encouraged to approach this analysis.
Further research could use samples of spontaneous or semi-spontaneous phrases and speech, along with habitual and disguised text reading, to obtain estimate/cutoff values in Hz from one measure to another. Thus, it would be possible to establish an acceptable Hz parameter in the future to determine whether the samples are believed to be from the same subject, making more in-depth inferences.
Studies whose methodological designs include accuracy, sensitivity, and specificity measures may enable the differentiation between individuals and speakers, in addition to the intra-individual comparisons carried out in this research.
CONCLUSION
The study results lead to the conclusion that fb, minimum f0, and maximum f0 were more stable in disguised speech, suggesting future usefulness in intra-individual comparisons. Hence, fb, a measure still little explored and disseminated in criminal forensics, proves useful for the comparative analysis of speakers, assisting in the production of the expert report. The speakers’ various types of adjustments to disguise their voices did not significantly alter fb, maximum f0, and minimum f0 measures in the comparison between habitual and disguised speech.
-
Study conducted at Universidade Estadual do Centro-Oeste – UNICENTRO - Irati (PR), Brasil.
-
Financial support:
Scientific Initiation Scholarship granted by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), call 2020/2021 of the Universidade Estadual do Centro-Oeste.
-
Data Availability:
Research data is not available.
-
Use of artificial intelligence-assisted technology
The authors declare that no artificial intelligence tools were used in the research reported here or in the preparation of this article.
References
-
1 Silva RRD, Costa JPCL, Miranda RK, Del Grado G. Aplicação do valor de base da frequência fundamental via estatística MVKD em comparação forense de locutor. Rev Bras Criminal. 2016;5(3):30-8. https://doi.org/10.15260/rbc.v5i3.134
» https://doi.org/10.15260/rbc.v5i3.134 -
2 Silva AP, Cantoni MM. Comparação forense de locutor por modelos lineares generalizados de variabilidade articulatória e vocal na fala encadeada. Avante. Rev Acad Polícia Minas Gerais. 2024;1(7):1-25. https://doi.org/10.70365/2764-0779.2024.101
» https://doi.org/10.70365/2764-0779.2024.101 -
3 Rose P. Forensic speaker identification. London: Taylor & Francis; 2002. https://doi.org/10.1201/9780203166369
» https://doi.org/10.1201/9780203166369 - 4 Rehder MI, Cazumbá LAF, Cazumbá MA. Identificação de falantes: uma introdução à fonoaudiologia forense. In: Sanches AP, Cazumbá LAF, Telles IFC, editores. Introdução à fonoaudiologia forense. 1. ed. Rio de Janeiro: Revinter; 2015. p. 7-24.
-
5 Chango X, Flor-Unda O, Gil-Jiménez P, Gómez-Moreno H. Technology in forensic sciences: innovation and precision. Technologies. 2024;12(8):120. https://doi.org/10.3390/technologies12080120
» https://doi.org/10.3390/technologies12080120 -
6 Carmo S, Rehder MIBC, Almeida LN, Villegas C, Dantas CRV, Vasconcelos D, et al. Forensic analysis of auditorily similar voices. Rev CEFAC. 2023;25(2):e4022. https://doi.org/10.1590/1982-0216/20232524022
» https://doi.org/10.1590/1982-0216/20232524022 - 7 Silva AP, Travassos AA, Ribeiro SMC. O estabelecimento de um corpus forense para o estudo de particularidades do dialeto mineiro. In: VIII Seminário Nacional de Fonética Forense; 2009; Palmas, TO. Anais. Brasília: Associação Brasileira de Criminalística; 2009.
-
8 Fulop SA, Disner SF. Advanced time-frequency displays applied to forensic speaker identification. Proc Meet Acoust. 2009;6(1):060008. https://doi.org/10.1121/1.3277007
» https://doi.org/10.1121/1.3277007 -
9 Gold E, French P. International practices in forensic speaker comparisons. Int J Speech Lang Law. 2019;26(1):1-20. https://doi.org/10.1558/ijsll.38028
» https://doi.org/10.1558/ijsll.38028 - 10 Nolan F. The phonetic bases of speaker recognition. Cambridge: Cambridge University Press; 1983.
-
11 Künzel HJ. Effects of voice disguise on speaking fundamental frequency. Int J Speech Lang Law. 2000;7(2):149-79. https://doi.org/10.1558/sll.2000.7.2.149
» https://doi.org/10.1558/sll.2000.7.2.149 -
12 Silva A. Dados de referência de F0 em corpus de falantes do Português Brasileiro na variedade falada na Capital Paulista. Rev Bras Criminal. 2022;11(2):92-105. https://doi.org/10.15260/rbc.v11i2.612
» https://doi.org/10.15260/rbc.v11i2.612 -
13 Wulf AN, Cruz PJA, Rosa BCS, Santos TD, César CPHAR. Tools and protocols used in criminal skills related to voice: literature review. Disturb Comun. 2020;32(1):52-63. https://doi.org/10.23925/2176-2724.2020v32i1p52-63
» https://doi.org/10.23925/2176-2724.2020v32i1p52-63 -
14 Hansen JH, Hasan T. Speaker recognition by machines and humans: a tutorial review. IEEE Signal Process Mag. 2015;32(6):74-99. https://doi.org/10.1109/MSP.2015.2462851
» https://doi.org/10.1109/MSP.2015.2462851 - 15 Kremer RL, Gomes ML. A eficiência do disfarce em vozes femininas: uma análise da frequência fundamental. ReVEL. 2014;12(23):28-34.
-
16 Lindh J, Eriksson A. Robustness of long time measures of fundamental frequency. In: 8th Annual Conference of the International Speech Communication Association (INTERSPEECH); 2007 Aug 27-31; Antwerp, Belgium. Proceedings. Brussels: ISCA; 2007. p. 2025-8. https://doi.org/10.21437/Interspeech.2007-166
» https://doi.org/10.21437/Interspeech.2007-166 -
17 Arantes P, Eriksson A. Temporal stability of long-term measures of fundamental frequency. In: 7th International Conference on Speech Prosody; 2014; Dublin. Proceedings. Brussels: ISCA; 2014. https://doi.org/10.21437/SpeechProsody.2014-220
» https://doi.org/10.21437/SpeechProsody.2014-220 -
18 Traunmüller H, Eriksson A. The frequency range of the voice fundamental in the speech of male and female adults [Internet]. 1995 [citado em 2025 Set 12]. 11 p. Disponível em: http://www2.ling.su.se/staff/hartmut/f0_m&f.pdf
» http://www2.ling.su.se/staff/hartmut/f0_m&f.pdf -
19 Traunmüller H. Conventional, biological and environmental factors in speech communication: a modulation theory. Phonetica. 1994;51(1-3):170-83. https://doi.org/10.1159/000261968 PMid:8052672.
» https://doi.org/10.1159/000261968 -
20 Masthoff H. A report on a voice disguise experiment. Int J Speech Lang Law. 1996;3(1):160-7. https://doi.org/10.1558/ijsll.v3i1.160
» https://doi.org/10.1558/ijsll.v3i1.160 -
21 Paula Machado A, Barbosa PA. Uso de técnicas acústicas para verificação de locutor em simulação experimental. Ling Direito. 2014;1(2):100. https://doi.org/10.47749/T/UNICAMP.2014.941972
» https://doi.org/10.47749/T/UNICAMP.2014.941972 -
22 Eriksson A. The disguised voice: imitating accents or speech styles and impersonating individuals. Lang Identity. 2010;8:86-96. https://doi.org/10.1515/9780748635788-012
» https://doi.org/10.1515/9780748635788-012 -
23 Cavalcanti JC, Eriksson A, Barbosa PA. Multiparametric analysis of speaking fundamental frequency in genetically related speakers using different speech materials: some forensic implications. J Voice. 2024;38(1):243.e11-29. https://doi.org/10.1016/j.jvoice.2021.08.013 PMid:34629229.
» https://doi.org/10.1016/j.jvoice.2021.08.013 - 24 Gomes MLC, Carneiro DO. A fonética forense no Brasil: cenários e atores. Ling Direito. 2014;1(1):22-36.
- 25 Giller R. O disfarce da voz em fonética forense [dissertação]. Lisboa: Universidade de Lisboa; 2011.
-
26 Gomes MLC, Carneiro DO, Dresch AAG. Análise perceptiva e acústica em fonética forense. Domínios de Linguagem. 2016;10(2):559-89. https://doi.org/10.14393/DL22-v10n2a2016-7
» https://doi.org/10.14393/DL22-v10n2a2016-7 -
27 Boss D. The problem of F0 and real-life speaker identification: a case study. Forensic Linguistics. 1996;3(1):155-9. https://doi.org/10.1558/ijsll.v3i1.155
» https://doi.org/10.1558/ijsll.v3i1.155 -
28 Graddol D, Swann J. Speaking fundamental frequency: some physical and social correlates. Lang Speech. 1983;26(Pt 4):351. https://doi.org/10.1177/002383098302600403 PMid:6677829.
» https://doi.org/10.1177/002383098302600403 - 29 Braid ACM. Fonética forense: tratado de perícias criminalísticas. Campinas: Millenium; 2003.
-
30 Mathur S, Chouldhary SK, Vyas JM. Effect of disguise on fundamental frequency of voice. J Forensics Res. 2016;7(3):3. https://doi.org/10.4172/2157-7145.1000327
» https://doi.org/10.4172/2157-7145.1000327 - 31 Lindh J. Preliminary F0 statistics and forensic phonetics. In: 15th Conference of the International Association for Forensic Phonetics and Acoustics; 2006; Göteborg, Sweden. Proceedings. Montreal: IAFPA; 2006.
Edited by
-
Editor:
Ana Carolina Constantini.
Research data is not available.
