一种机器人智能语音识别算法研究

针对智能机器人在非特定人语音识别中识别率偏低的问题,提出了一种双门限的端点检测算法,精确地检测出了语音端点,对分形维数和Mel频率倒谱系数(MFCC)进行结合,同时基于隐马尔可夫(HMM)模型,提出了智能机器人命令识别系统;在实验室环境下,利用Cool Edit软件录制了5男5女的语音,采样率为8 kHz,精度为16位,内容为5个命令词,每个词均被采集6次,将每人的前3次发音作为模板语音,后3次发音作为测试语音,实验结果表明,系统识别率可以达到85%以上,MFCC与分形维数混合的语音特征参数的算法提高了系统识别率,优化了系统性能;该方法用于非特定人语音智能识别是可行的、有效的。     

Outcome of laryngeal manual therapy in four Dutch adults with persistent moderate-to-severe vocal hyperfunction: a pilot study

A relatively new management strategy for the treatment of voice disorders is the use of laryngeal manual therapy. The main purpose of the present pilot study is to document the outcome of vocal quality after a well-defined laryngeal manual therapy (LMT) program. Four Dutch professional voice users with a persistent moderate or severe muscle tension dysphonia were studied pretreatment (1 week before LMT) and posttreatment (1 week) after completion of manual therapy (25 sessions). These subjects had received several months of traditional voice therapy, without any success. To measure and compare, the effect of LMT objective and subjective assessment techniques were used. Perceptual voice assessment included a perceptual rating of the voice using the GRBAS scale. Furthermore, the vocal quality in this population was modeled by means of the Dysphonia Severity Index (DSI). All of the subjects selected for LMT showed improvement in perceptual vocal quality and DSI values. As the DSI is a weighted variable including aerodynamic and acoustic measures, small improvements (closer to 5) are very indicative of vocal quality improvement.¹ The use of LMT in professional voice users with persistent moderate-to-severe muscle tension dysphonia, especially in some subjects who have not responded to traditional voice therapy, is supported by this pilot study.     

Measurement of Characteristic Leap Interval Between Chest and Falsetto Registers

A markedly smaller time constant distinguishes a chest-falsetto leap from the more usual execution of a sung interval by muscular adjustments in the length and tension of the vocal folds. The features of such a chest-falsetto leap are examined in detail with respect to F0, peak-to-peak amplitude of the vocal-fold contact area signal (EGG), and the closed quotient. A method is proposed to standardize and quantify this chest-falsetto leap in the characteristic leap interval (CLI), a measure of the separation between the natural registers in a given singing voice. The measure is applied to a varied group of experienced singers. Preliminary results include a suggested dimorphic pattern with respect to sex, with female voices exhibiting smaller CLIs and less individual diversity than male voices.     

Poor Voice Quality in Future Elite Vocal Performers and Professional Voice Users

The voice quality of 86 occupational voice users, i.e., students of a high school for audiovisual communication, was assessed by means of a multidimensional test battery containing: the GRBAS scale, videolaryngostroboscopy, maximum phonation time, jitter, lowest intensity, highest frequency, dysphonia severity index (DSI), and voice handicap index (VHI). In a questionnaire on daily habits the prevalence of smoking, eating habits, and vocal abuse were recorded. A comparison of the voice characteristics of the future occupational voice users with a control group revealed significant differences. The results of the VHI and the DSI of these students revealed significantly worse scores than the score of a control group characterized by no vocal complaints. Moreover, the questionnaire on daily habits showed that the future elite vocal performers and professional voice users take less precautions for the care of their voices. These findings support the importance of a good balanced vocal coaching.     

Acoustic Analyses of Trained Singers Perceptually Identified from Speaking Samples

This study investigated selected acoustic cues in the speaking voices of five professional singers; cues that may have enabled na?ve listeners to differentiate them from nonsingers and other trained singers who were not consistently identified from their speaking voices. Subjects were divided into three groups based on listeners' perceptual judgments. Group I, the identified singers, consisted of five professional singers, three males and two females, with an average identification score, from their speaking utterances, of 79%. Group II, the unidentified singers, consisted of 15 trained singers, seven males and eight females, who, as a group, were identified correctly from their speaking utterances only 52% of the time. Group III consisted of 20 nonsingers who were incorrectly identified from their speaking utterances as singers only 36% of the time, that is, they were correctly identified as nonsingers from their speech 64% of the time. Acoustic parameters chosen for measurement from vowel productions were: (1) percent jitter, (2) percent shimmer, and (3) noise-to-harmonic ratio. The second sentence of the "Rainbow Passage" was selected to compare several frequency and duration measures between the three groups. These were: (1) mean speaking fundamental frequency, (2) standard deviation of the fundamental frequency, (3) sentence duration, (4) word duration, and (5) consonant/vowel ratio. The data indicated that the acoustic parameters that most consistently distinguished the identified singers from the unidentified singers and the nonsingers were fundamental frequency variation and durational differences. The identified singers varied their speaking fundamental frequency significantly more than did both the unidentified singers and the nonsingers. The identified singers also had longer vocalic segments than did the others.     

Long-Term-Average Spectrum Characteristics of Country Singers During Speaking and Singing

Five premier male country singers involved in our previous studies spoke and sang the words of both the national anthem and a country song of their choice. Long-term-average spectra were made of the spoken and sung material of each singer. The spectral characteristics of county singers' speech and singing were similar. A prominent peak in the upper part of the spectrum, previously described as the "speaker's formant," was found in the county singers' speech and singing. The singer's formant, a strong spectral peak near 2.8 kHz, an important part of the spectrum of classically trained singers, was not found in the spectra of the country singers. The results support the conclusion that the resonance characteristics in speech and singing are similar in country singing and that county singing is not characterized by a singer's formant.     

Resonance strategies for the belting style: Results of a single female subject study

This study explores resonance strategies used for the belting style and associated vocal fold vibratory patterns, for the vowels /e/, /a/, /i/, and /u/ on G4 and B4-flat. Acoustic spectra of belted vowels and their unoptimized, "speech-like" equivalents were compared. Vocal fold vibratory patterns were quantified using electroglottography. Results show that /a/ is inherently suitable for belting and requires no adjustment. For /e/, F2-H5 tuning was observed. For /i/, F1 was detuned from H1, enhancing also H2. For /u/, both F1 and F2 were raised to accomplish F2-H3 tuning. These results show that the loud, bright sound of the belting style is achieved by the implementation of resonance strategies that enhance higher harmonics. Electroglottography revealed that resonance strategies also result in raising the closed quotient (CQ) above 52%, an apparent threshold value for belting.     

Perceptual confusions of high-pitched sung vowels

Questions exist as to the intelligibility of vowels sung at extremely high fundamental frequencies and, especially, when the fundamental frequency (F0) produced is above the region where the first vowel formant (F1) would normally occur. Can such vowels be correctly identified and, if so, does context provide the necessary information or are acoustical elements also operative? To this end, 18 professional singers (5 males and 13 females) were recorded when singing 3 isolated vowels at high and low pitches at both loud and soft levels. Aural-perceptual studies employing four types of auditors were carried out to determine the identity of these vowels, and the nature of the confusions with other vowels. Subsequent acoustical analysis focused on the actual fundamental frequencies sung plus those defining the first 2 vowel formants. It was found that F0 change had a profound effect on vowel perception; one of the more important observations was that the target tended to shift toward vowels with an F1 just above the sung frequency.     

Effects of vocal training on the acoustic parameters of the singing voice

Vocal training (VT) has, in part, been associated with the distinctions in the physiological, acoustic, and perceptual parameters found in singers' voices versus the voices of nonsingers. This study provides information on the changes in the singing voice as a function of VT over time. Fourteen college voice majors (12 females and 2 males; age range, 17–20 years) were recorded while singing, once a semester, for four consecutive semesters. Acoustic measures included fundamental frequency (F₀) and sound pressure level (SPL) of the 10% and 90% levels of the maximum phonational frequency range (MPFR), vibrato pulses per second, vibrato amplitude variation, and the presence of the singer's formant. Results indicated that VT had a significant effect on the MPFR. F₀ and SPL of the 90% level of the MPFR and the 90–10% range increased significantly as VT progressed. However, no vibrato or singers' formant differences were detected as a function of training. This longitudinal study not only validates previous cross-sectional research, ie, that VT has a significant effect on the singing voice, but also it demonstrates that these effects can be acoustically detected by the fourth semester of college vocal training.     

应用ResNet和CatBoost检测重放语声*

针对短语声指令声频信息少、不适用句子级重放语声检测的问题以及近距离录声后用高质量重放设备重放的语声难以检测的问题，提出了一种适用于词级重放语声检测的模型。首先，利用短时傅里叶变换、低频平均能量计算和帧排序等方法选择声频帧，然后提取这些帧的伽马通频率倒谱系数。其次，用基于自注意机制的残差网络模型进一步提取伽马通频率倒谱系数中的信息，并转化为特征向量。最后，将提取后的特征向量用CatBoost分类，从而提高检测性能。在POCO数据集上的实验结果表明，提出的方法可以以87.54%的准确率和12.53%的等错误率检测重放语声，优于基线和现有的方法。该文提出的方法在ASVspoof2019 PA数据集上的等错误率与串联检测代价函数分别为4.92%和0.1418，证明该文方法也适用于多种设置的重放语声检测。