Voice Conversion

The basic goal of the voice conversion system to mimics the characteristics of the target speaker voice by keeping the linguistic and paralinguistic information intact. The characteristics of a speaker in speech reflect at different level such as vocal tract, excitation and prosodic parameters. This propose work based on cepstrum which represents the vocal tract and excitation parameters of the speech. This paper proposes the decomposition of the cepstrum by wavelet and mapped the source cepstrum features in to target cepstrum features using Radial

The basic goal of the voice conversion system to mimics the characteristics of the target speaker voice by keeping the linguistic and paralinguistic information intact. The characteristics of a speaker in speech reflect at different level such as vocal tract, excitation and prosodic parameters. This propose work based on cepstrum which represents the vocal tract and excitation parameters of the speech. This paper proposes the decomposition of the cepstrum by wavelet and mapped the source cepstrum features in to target cepstrum features using Radial basis function neural network. The results are evaluated using subjective and objective measures based on voice quality method and the listening tests prove that the proposed algorithm converts speaker individuality while maintaining high speech quality..

The basic goal of the voice conversion system to mimics the characteristics of the target speaker voice by keeping the linguistic and paralinguistic information intact. The characteristics of a speaker in speech reflect at different level such as vocal tract, excitation and prosodic parameters. This propose work based on cepstrum which represents the vocal tract and excitation parameters of the speech. This paper proposes the decomposition of the cepstrum by wavelet and mapped the source cepstrum features in to target cepstrum features using Radial basis function neural network. The results are evaluated using subjective and objective measures based on voice quality method and the listening tests prove that the proposed algorithm converts speaker individuality while maintaining high speech quality..

Recent advancements in voice conversion systems have been largely driven by deep learning techniques, enabling the high-quality synthesis of human speech. However, existing models often fail to generate emotionally expressive speech, resulting in outputs that lack the richness of human communication. The main goal of this research is to fill this gap by incorporating a prosody-aware extension into embedding-guided neural voice conversion (EGNVC). This extension seeks to improve the emotional expressiveness and naturalness of the converted speech, while preserving both speaker identity and linguistic content. A new framework is introduced that integrates a content encoder, speaker embedder, and prosody extractor to address this challenge. The prosody extractor is designed to capture dynamic prosodic elements like pitch, energy, and timing from a reference audio signal. These features are then introduced into the decoding process through a prosody conditioning module, allowing for precise control over prosody during speech synthesis. FiLM (Feature-wise Linear Modulation) layers are employed to adjust intermediate features based on speaker and prosody maintained. The vocoder converts the generated mel-spectrograms into high-quality waveforms, ensuring a natural-sounding output. Experimental results show that the proposed model surpasses conventional voice conversion systems in maintaining linguistic content and prosodic features, with significant improvements in Mel-Cepstral Distortion (MCD), Pitch RMSE, and Energy RMSE. Subjective assessments, including Mean Opinion Score (MOS) and Emotion Classification Accuracy, highlight the model's enhanced emotional expressiveness and naturalness. These findings emphasize the critical role of prosody in high-quality voice conversion, offering a robust framework for creating more dynamic, expressive, and human-like speech synthesis systems.

Two speech processing systems have been developed for real-time and non-real-time voice conversion. Using the real-time processing the user can apply conversion during voice over IP (VoIP) calls imitating identity of a specified target speaker. Non-real-time processing system converts prerecorded audio books read by a professional reader imitating voice of the user. Both systems require some speech samples of the user for training. The training procedures are similar for both systems however the user is considered as a source speaker in the first case and as a target speaker in the second. For parametric representation of speech we use a speech model based on instantaneous harmonic parameters with multicomponent sinusoidal excitation. The voice conversion itself is made using artificial neural networks (ANN) with rectified linear units. Here we demonstrate implementations of the voice conversion systems with dedicated web interfaces and iPhone application.

A real-time pitch modification system has been developed. The implemented processing scheme is based on hybrid deterministic/stochastic decomposition of the signal and includes extraction of instantaneous pitch, pitch-synchronous time-frequency analysis, parametrical morphing and synthesis. The scheme provides high quality output with considerably high naturalness. The aim of the presentation is to show capabilities of the designed real-time signal processing framework. The system implements speech-specific intonation change routines such as lowering, uplifting, tremolo etc. In order to make the presentation more expressive we designed a special singing mode in which the system automatically corrects wrong notes. The target melody and voice effects are specified using musical instruments digital interface (MIDI).

Radio frequencies for controller pilot communication are becoming a scarce resource due to increasing air traffic worldwide. The controller pilot data link communication (CPDLC) technology, which is already mandatory in new aircrafts, aims at reducing radio interferences and misunderstandings of commands by allowing routine voice conversations to be carried out via text messaging. This paper describes the requirements, architecture, and development of an air traffic management (ATM) system with CPDLC capabilities in ...

The NAM-to-speech conversion proposed by Toda and colleagues which converts Non-Audible Murmur (NAM) to audible speech by statistical mapping trained using aligned corpora is a very promising technique, but its performance is still insufficient. In this paper, we present our current work to improve the intelligibility and the naturalness of the synthesized speech converted from whispered speech with this technique. The first system is proposed to improve F 0 estimation and voicing decision. A simple neural network is used to detect voiced segments in the whisper while a GMM estimates a continuous melodic contour based on training voiced segments. In the second system, we attempt to integrate visual information for improving both spectral estimation, F 0 estimation and voicing decision.

In this paper, we describe a concatenative synthesis sys- tem which was first designed for a realistic synthesis of melodic phrases. It has since been augmented to become an experimental TTS (Text-to-Speech) synthesizer. To- day, it is able to realize hybrid synthesis involving speech segments and musical excerpts coming from any record- ing imported in its database. The system can also synthe- size sentences with different voices, sentences with mu- sical sounds, melodic phrases with speech segments and generate compositional material from specific intonation patterns using a prosodic pattern extractor.

Spectral envelopes are very useful in sound analysis and synthesis because of their connection with production and perception models, and their ability to capture and to manipulate important properties of sound using easily understandable "musical" parameters. It is not easy, however, to estimate and represent them well, as several requirements must be fulfilled. We discuss the strengths and weaknesses of the estimation methods LPC, cepstrum, and discrete cepstrum, and evaluate the representations filter coefficients, sampled, break-point functions, splines, and formants. The proposed high-level approach to spectral envelope handling is followed in software developed at IRCAM, which makes some important applications of spectral envelopes in the domain of additive analysis-synthesis possible.

The recent explosive growth of voice over IP (VoIP) solutions calls for accurate modelling of VoIP traffic. This paper presents measurements of ON and OFF periods of VoIP activity from a significantly large database of VoIP call recordings consisting of native speakers speaking in some of the world's most widely spoken languages. The impact of the languages and the varying dynamics of caller interaction on the ON and OFF period statistics are assessed. It is observed that speaker interactions dominate over language dependence which makes monologue based data unreliable for traffic modelling. The authors derive a semi-Markov model which accurately reproduces the statistics of composite dialogue measurements.

The recent explosive growth of voice over IP (VoIP) solutions calls for accurate modelling of VoIP traffic. This paper presents measurements of ON and OFF periods of VoIP activity from a significantly large database of VoIP call recordings consisting of native speakers speaking in some of the world's most widely spoken languages. The impact of the languages and the varying dynamics of caller interaction on the ON and OFF period statistics are assessed. It is observed that speaker interactions dominate over language dependence which makes monologue based data unreliable for traffic modelling. The authors derive a semi-Markov model which accurately reproduces the statistics of composite dialogue measurements.

This paper describes the implementation of a unit selection text-to-speech system that incorporates a statistical model Cost (sCost), in addition to target and join costs, for controlling the selection of unit candidates. sCost, a quality control measure, is calculated off-line for each unit by comparing HMM based synthesis and recorded speech with their corresponding unit segment labels. Dynamic time warping (DTW) is used to perform such comparison at level of spectrum, pitch and voice strengths. The method has been tested on unit selection voices created using audio book data. Preliminary results indicate that the use of sCost based only on spectrum introduce more variety on style pronunciation but affects quality; whereas using sCost based on spectrum, pitch and voicing strengths improves significantly the quality, maintaining a more stable narrative style.

Este trabajo explora el uso de inteligencia artificial en la creación de recursos para el ejercicio de la comprensión auditiva en la enseñanza del griego antiguo como lengua hablada. Partiendo de la importancia de la adquisición de esta destreza para un aprendizaje efectivo de la lengua, este proyecto busca suplir la ausencia de recursos y dotar al docente de herramientas para la generación autónoma de materiales.
Los resultados de su puesta en práctica durante un taller piloto evidencian una recepción muy positiva por parte de los participantes y una mejora significativa en su comprensión auditiva. En este sentido, la metodología propuesta puede ser aplicada no solo a otras lenguas clásicas como el latín, sino también a cualquier lengua moderna.

A method for modifying voice quality attributes, i.e. breathiness and roughness, is presented in the context of voice conversion. Both breathiness and roughness of a speaker are collectively modelled by harmonic peak-to-valley ratio (HPVR) in the speaker's speech spectrum. The average HPVR is modified through a post-filtering operation after the conversion of spectral envelope, pitch and other speaker individuality features. It is shown that average HPVR for various speakers correlates well with the perceived breathiness and roughness in their voices. Further, it is shown through subjective experiments that the proposed modification in HPVR results in improved subjective rating of the transformed speech signal.

This paper presents a novel approach for enhancing esophageal speech using voice conversion techniques. Esophageal speech (ES) is an alternative voice that allows a patient with no vocal cords to produce sounds after total laryngectomy: this voice has a poor degree of intelligibility and a poor quality. To address this issue, we propose a speaking-aid system enhancing ES in order to clarify and make it more natural. Given the specificity of ES, in this study we propose to apply a new voice conversion technique taking into account the particularity of the pathological vocal apparatus. We trained deep neural networks (DNNs) and Gaussian mixture models (GMMs) to predict "laryngeal" vocal tract features from esophageal speech. The converted vectors are then used to estimate the excitation cepstral coefficients and phase by a search in the target training space previously encoded as a binary tree. The voice resynthesized sounds like a laryngeal voice i.e., is more natural than the original ES, with an effective reconstruction of the prosodic information while retaining, and this is the highlight of our study, the characteristics of the vocal tract inherent to the source speaker. The results of voice conversion evaluated using objective and subjective experiments, validate the proposed approach.

Dans cet article, nous présentons un système de Reconnaissance Automatique de la Parole (RAP) combinant les données acoustiques et les données visuelles. Ce système de reconnaissance audiovisuelle utilise comme moteur de reconnaissance les modèles de Markov cachés continus (Continuous Hidden Markov Model, CHMM) et comme méthode de fusion l'Identification Séparée (IS) basée sur les réseaux de neurones de type Perceptron Multi Couches (PMC). L'information visuelle utilisée conjointement avec les données acoustiques est basée sur la forme et les mouvements des lèvres lors de l'élocution. Les expériences réalisées pour la reconnaissance des chiffres arabes indiquent que l'utilisation conjointe de la modalité acoustique et de la modalité visuelle augmente la performance du système de RAP en milieu réel (fortement bruité), une augmentation du Taux ce Reconnaissance (TR) de l'ordre de 15% a été constatée.

The paper presents and evaluates a speaker de-identification technique using speech recognition and two speech synthesis techniques. The phoneme recognition system is built using HMM-based acoustical models of contextdependent diphone speech units, and two different speech synthesis systems (diphone TD-PSOLA-based and HMM-based) are employed for re-synthesizing the recognized sequences of speech units. Since the acoustical models of the two speech synthesis systems are assumed to be completely independent of the input speaker's voice, the highest level of input speaker de-identification is ensured. The proposed de-identification system is considered to be language dependent, but is, however, vocabulary and speaker independent since it is based mainly on acoustical modelling of the selected diphone speech units. Due to the relatively simple computing methods, the whole de-identification procedure runs in realtime. The speech outputs are compared and assessed by testing the intelligibility of the re-synthesized speech from different points of view. The assessment results show interesting variabilities of the evaluators' transcriptions depending on the input speaker, the synthesis method applied and the evaluators capabilities. But in spite of the relatively high phoneme recognition error rate (approx. 19%), the re-synthesized speech is in many cases still fully intelligible.

Memorandum 4152 vibI1:,Cd JAvail 2nd/eor Dit Special. EXP)LICIT MODELLING OF STATE DURATION CORRELATIONS VN TMIDDEN MARKOV MODELS-5 M J Russell and L Siine September 1988 ABSTRACT4 fn recent years considerable effort has been directed towards impro'ving thc treatment of durational structure in hidden Markov model (1-MM) based approaches to speech pattern modelling. In general these studies have been concerned with more accurate modelling of. thc variations in segment duration which occur when words are spoken at a nominally constant speaking rate. However, recent work has shown that some of the performancc gains which can be achieved by improved duration modelling are lost when thc wor2s inl the test set are spoken at a different rate from those in the training set. This memorandum presents an approach to 5ol, ng this problem based on the capture and use of informa tion about state duration correlations. A mnethod for measuring correlations Ibetween the durations of 6djacent states in a HMM is described. The method involves expanding a standard HMM into a special type of hidden se-ni-NMarkov model (HSMM;\), caled a Correlated Duration H-MM (CDHMIM), in which each state of the original HMNM is expanded into a set of fixed-duration HSMM states. The probabilities associatce '-ith transitions between these states are measures of state duration correlaion. Experiments ar~c described in which the CDHMM method is applied to a set of sentences spoken at four different speaking rates. Note: This memorandum is an expanded version of a paper trescntcd Pt SPEECH-8'SS Sevcn-,h FASE.-Symnposiumn, 22-26 August 1958, Edinburgh. Copyright 0Controier tMSZO, London, 198&. I ~~ I I..

HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

In this paper, new techniques to improve whisper-to-speech conversion are investigated, in the framework of silent speech telephone communication. A preliminary conversion method from Non-Audible Murmur (NAM) to modal speech, based on statistical mapping trained using aligned corpora has been proposed. Although it is a very promising technique, its performance is still insufficient due to the difficulties in estimating F 0 from unvoiced speech. In this paper, two distinct modifications are proposed, in order to improve the naturalness of the synthesized speech. In the first modification, LDA (Linear Discriminant Analysis) is used instead of PCA (Principal Component Analysis) to reduce the dimensionality of the input spectral vectors. In addition, the influence of long-term variation of spectral information on pitch estimation is examined. The second modification is an attempt to integrate visual information as a complementary input to improve spectral estimation, F 0 estimation and voicing decision.

The compact representation of harmonic amplitudes in the sinusoidal coding of voiced speech is often achieved by the all-pole modeling of a spectral envelope. The perceptual accuracy of the representation may be enhanced by the use of frequency-scale warping prior to LP modeling. The effectiveness of a frequency warping function has been shown in previous work to depend on the underlying signal spectral shape as determined by phoneme quality as well as voice quality. This suggests the possibility of minimizing the perceived degradation due to LP modeling by adapting the warping function to the local signal spectrum. In this paper, an investigation is conducted to determine possible correlations between preferred warping condition and easily computable spectral cues. Experimental results are presented that demonstrate the advantage of adapting the warping function to the signal spectral envelope in the context of a sinusoidal speech coding scheme.

This paper presents a pitch modification scheme, based on the recursive least-squares (RLS) adaptive algorithm, for speech and singing voice signals. The RLS filter is used to determine the linear prediction (LP) model on a sample-by-sample framework, as opposed to the LP-coding (LPC) method, which operates on a block basis. Therefore, an RLS-based approach is able to preserve the natural subtle variations on the vocal tract model, avoiding discontinuities in the synthesized signal and the inherent frame-delay associated to classic methods. The LP residual is modified in the synthesis stage in order to generate the output signal. Listening tests verify the overall quality of the synthesized signal using the RLS approach, indicating that this technique is suitable for realtime applications.

Dans le domaine scientifique, le stockage et la gestion efficaces des données sont cruciaux pour la recherche, l'analyse et la collaboration. Différents formats de fichiers ont été développés pour répondre aux besoins spécifiques des diverses disciplines scientifiques. Voici une introduction aux cinq formats de stockage de données les plus couramment utilisés : DICOM, NetCDF, HDF5, FITS et CSV.

This paper presents a novel approach for enhancing esophageal speech using voice conversion techniques. Esophageal speech (ES) is an alternative voice that allows a patient with no vocal cords to produce sounds after total laryngectomy: this voice has a poor degree of intelligibility and a poor quality. To address this issue, we propose a speaking-aid system enhancing ES in order to clarify and make it more natural. Given the specificity of ES, in this study we propose to apply a new voice conversion technique taking into account the particularity of the pathological vocal apparatus. We trained deep neural networks (DNNs) and Gaussian mixture models (GMMs) to predict "laryngeal" vocal tract features from esophageal speech. The converted vectors are then used to estimate the excitation cepstral coefficients and phase by a search in the target training space previously encoded as a binary tree. The voice resynthesized sounds like a laryngeal voice i.e., is more natural than the original ES, with an effective reconstruction of the prosodic information while retaining, and this is the highlight of our study, the characteristics of the vocal tract inherent to the source speaker. The results of voice conversion evaluated using objective and subjective experiments, validate the proposed approach.

This paper describes the evaluation process of an emotional speech database recorded for standard Basque in order to determine its adequacy for the analysis of emotional models and its use in speech synthesis. The corpus consists of seven hundred semantically neutral sentences ...

In this paper, we propose a real-time method for duration modification of speech for packet based communication system. While there is rich literature available on duration modification, it fails to clearly address the issues in real-time implementation of the same. Most of the duration modification methods rely on accurate estimation of pitch marks, which is not feasible in a real-time scenario. The proposed method modifies the duration of Linear Prediction residual of individual frames without using any look-ahead delay and knowledge of pitch marks. In this method, multiples of pitch period is repeated or removed from a frame depending on a scheduling algorithm. The subjective quality of the proposed method was found to be better than waveform similarity overlap and add (WSOLA) technique as well as Linear Prediction Pitch Synchronous Overlap and Add (LP-PSOLA) technique.

The evaluation procedure to choose a low bit rate voice coding algorithm is described for the Australian land mobile satellite system. The procedure is designed to assess both the inherent quality of the codec under 'normal' conditions and its robustness under 'severe' conditions. For the assessment, normal conditions were chosen to be random bit error rate with added background acoustic noise and the severe condition is designed to represent burst error conditions when mobile satellite channel suffers from signal fading due to roadside vegetation. The assessment is divided into two phases. First, a reduced set of conditions is used to determine a short list of candidate codecs for more extensive testing in the second phase. The first phase conditions include quality and robustness and codecs are ranked with a 60:40 weighting on the two. Second, the short listed codecs are assessed over a range of input voice levels, BERs, background noise conditions, and burst e...

This paper addresses a phase-related feature that is time-shift invariant, and that expresses the relative phases of all harmonics with respect to that of the fundamental frequency. We identify the feature as Normalized Relative Delay (NRD) and we show that it is particularly useful to describe the holistic phase properties of voiced sounds produced by a human speaker, notably vowel sounds. We illustrate the NRD feature with real data that is obtained from five sustained vowels uttered by 20 female speakers and 17 male speakers. It is shown that not only NRD coefficients carry idiosyncratic information, but also their estimation is quite stable and robust for all harmonics encompassing, for most vowels, at least the first four formant frequencies. The average NRD model that is estimated using data pertaining to all speakers in our database is compared to that of the idealized Liljencrants-Fant (LF) and Rosenberg glottal models. We also present results on the phase effects of linear-...

Vowel recognition is frequently based on Linear Prediction (LP) analysis and formant estimation techniques. However, the performance of these techniques decreases in the case of female or child speech because at high pitch frequencies (F0) the magnitude spectrum is scarcely sampled making formant estimation unreliable. In this paper we describe the implementation of a perceptually motivated concept of vowel recognition that is based on Perceptual Spectral Clusters (PSC) of harmonic partials. PSC based features were evaluated in automatic recognition tests using the Mahalanobis distance and using a data base of five natural Portuguese vowel sounds uttered by 44 speakers, 27 of whom are child speakers. LP based features and Mel-Frequency Cepstral Coefficients (MFCC) were also included in the tests as a reference. Results show that while the recognition performance of PSC features falls between that of LP based features and that of MFCC coefficients, the normalization of PSC features by F0 increases the performance and approaches that of MFCC coefficients. PSC features are not only amenable to a psychophysical interpretation (as LP based features are) but have also the potential to compete with global shape features such as MFCCs.

Residual prediction is a technique that aims at recovering the spectral details of speech that was encoded using parameterizations as linear predictive coefficients. Example applications of residual prediction are hidden Markov modelbased speech synthesis or voice conversion. Our voice conversion experiments showed that only one of the seven compared techniques was capable of successfully converting the voice while achieving a fair speech quality (i.e. mean opinion score = 3).

Residual prediction is a technique that aims at recovering the spectral details of speech that was encoded using parameterizations as linear predictive coefficients. Example applications of residual prediction are hidden Markov modelbased speech synthesis or voice conversion. Our voice conversion experiments showed that only one of the seven compared techniques was capable of successfully converting the voice while achieving a fair speech quality (i.e. mean opinion score = 3).

This paper presents an empirical study of several policies for managing the effect of delay jitter on the playout of audio and video in computer-based conferences. The problem addressed is that of managing the fundamental tradeoff between display with low latency and display with few gaps. We describe a particular policy called queue monitoring which observes delay jitter over time and dynamically adjusts display latency in order to support low-latency conferences with an acceptable gap rate. Queue monitoring is evaluated by comparing it with two policies from the literature in a study based on measurements from a computer-based conferencing system. Our results show that queue monitoring performs as well or better than the other policies over the range of observed network loads. More importantly, we show that queue monitoring performs better on those network loads for which the other policies exhibit poor performance.

There is a great demand to assess video quality transmitted in real time over packet networks, and to make this assessment in real time too. Quality assessment is achieved using two types of methods: objective or subjective. Subjective methods give more reliable results (objective ones do not correlate well with human perception), but unfortunately, they are not suitable to real-time applications and are difficult to carry out. In this paper, we show how Artificial Neural Networks (ANN) can be used to mimic the way by which a group of human subjects assess video quality when this video is distorted by certain quality-affecting parameters (e.g. packet loss rate, loss distribution, bit rate, frame rate, encoded frame type, etc.). Our method can be used to measure in real time the subjective video quality with very good precision. In order to illustrate its applicability, we chose to assess the quality of video flows transmitted over IP networks and we carried out subjective quality tests for video distorted by variations of those parameters.

Unconstrained face recognition remain a challenging problem due to intra-class variations caused by occlusion, disguise, varying orientations, facial expressions, age variations and illumination in real circumstances...etc. the recognition rate of traditional face recognition algorithms would be very low in this conditions. To address this issue, we propose a non-linear extension to the sparse representation classifier adapted to real-world conditions that can be trained using single training sample. We conduct extensive experiments on AR dataset to verify the efficacy of the proposed method.

We propose a novel application based on acoustic-toarticulatory inversion (AAI) towards quality assessment of voice converted speech. The ability of humans to speak effortlessly requires coordinated movements of various articulators, muscles, etc. This effortless movement contributes towards naturalness, intelligibility and speaker's identity (which is partially present in voice converted speech). Hence, during voice conversion, the information related to speech production is lost. In this paper, this loss is quantified for male voice, by showing increase in RMSE error (upto 12.7 % in tongue tip) for voice converted speech followed by showing decrease in mutual information (I) (by 8.7 %). Similar results are obtained in case of female voice. This observation is extended by showing that articulatory features can be used as an objective measure. The effectiveness of proposed measure over MCD is illustrated by comparing their correlation with Mean Opinion Score (MOS). Moreover, preference score of MCD contradicted ABX test by 100 %, whereas proposed measure supported ABX test by 45.8 % and 16.7 % in case of female-to-male and male-to-female voice conversion systems, respectively.

Recently, major services provided by mobile communications systems are shifting from voice conversations to data communications over the Internet. There is a strong demand for increasing the data transmission rate. However, an important problem arises; larger peak transmit power is required as transmission rate becomes higher. In this paper, a virtual cellular concept is proposed to avoid this power problem and its transmit power and frequency efficiencies are evaluated by computer simulation to compare with that of the present cellular systems.

Speaker’s identity is the most crucial information exploited (implicitly) by an Automatic Speaker Verification (ASV) system. Numerous attacks can be obliterated simultaneously if privacy preservation is exercised for a speaker’s identity. The baseline of the Voice Privacy Challenge 2020 by INTERSPEECH uses the Linear Prediction (LP) model of speech, and McAdam’s coefficient for achieving speaker de-identification. The baseline approach focuses on altering only the pole angles using McAdam’s coefficient. However, from speech acoustics and digital resonator design, the radius of the poles is associated with various energy losses. The energy losses implicitly carry speaker-specific information during speech production. To that effect, the authors have brought fine-tuned changes in both pole angle and pole radius, resulting in 18.98% higher value of EER for Vctk-test-com dataset, and 5% lower WER for Libri-test dataset compared to the baseline. This means privacy-preservation is indeed ...

We present a multi-speaker formant synthesizer based on parameter concatenation. The user can choose among three speakers, two males and one female. The synthesizer stores all the parameters for the basic speaker and linear transformation functions to synthesized the other two. The complete database for one speaker consists of 455 parameterized units (diphones, triphones,...) and the parameters used are pitch, formants and bandwidths and source parameters (four parameters for the LF model, and glottal noise). To get the converted speaker we store a linear transformation function for each spectral stable segment of each unit. Preliminary results show that the quality of the synthesizer is very good and that this system can help us to study and understand the speaker variability problem.

Generally defined, speech modification is the process of changing certain perceptual properties of speech while leaving other properties unchanged. Among the many types of speech information that may be altered are rate of articulation, pitch and formant characteristics.Modifying the speech parameters like pitch, duration and strength of excitation by desired factor is termed as prosody modification. In this thesis prosody modifications for voice conversion framework are presented.Among all the speech modifications for prosody two things are important firstly modification of duartion and pauses (Time scale modification) in a speech utterance and secondly modification of the pitch(pitch scale modification).Prosody modification involves changing the pitch and duration of speech without affecting the message and naturalness.In this work time scale and pitch scale modifications of speech are discussed using two methods Time Domain Pitch Synchronous Overlapped-Add (TD-PSOLA) and epoch based approach.In order to apply desired speech modifications TD-PSOLA discussed in this thesis works directly on speech in time domian although there are many variations of TD-PSOLA.The epoch based approach involves modifications of LP-residual. Among the various perceptual properties of speech pitch contour plays a key role which defines the intonation patterns of speaker.Prosody modifications of speech in voice conversion framework involve modification of source pitch contour as per the pitch contour of target.In a voice conversion framework it requires prediction of target pitch contour. Mean/ variance method for pitch contour prediction is explored. Sinusoidal modeling has been successfully applied to a broad range of speech processing problems. It offers advantages over linear predictive modeling and the short-time Fourier transform for speech analysis/ synthesis and modification. The parameter estimation of sinusoidal modeling which permits flexible time and frequency scale voice modifications is presented. Speech synthesis using three models sinusoidal, harmonic and harmonic-plus-residual is discussed. vi

Generally defined, speech modification is the process of changing certain perceptual properties of speech while leaving other properties unchanged. Among the many types of speech information that may be altered are rate of articulation, pitch and formant characteristics.Modifying the speech parameters like pitch, duration and strength of excitation by desired factor is termed as prosody modification. In this thesis prosody modifications for voice conversion framework are presented. Among all the speech modifications for prosody two things are important firstly modification of duartion and pauses (Time scale modification) in a speech utterance and secondly modification of the pitch(pitch scale modification).Prosody modification involves changing the pitch and duration of speech without affecting the message and naturalness.In this work time scale and pitch scale modifications of speech are discussed using two methods Time Domain Pitch Synchronous Overlapped-Add (TD-PSOLA) and epoch b...