Dutoit, Thierry, Vicent Pagel, N icolas Pierret, Francois Bataille and Olivier Van der Vrecken . 1996. The MBROLA Project: Towards a Set of H igh-Quality Speech Synthesizers Free of Use for Non Commercial Pu rpose . In Proceedings of ICSLP96,vol.3, p1393-1396.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the limitations of using a commercial product without access to the source code, a new synthesizer was developed and named emoSyn (acronym for emotionsynthesizer) 2.1. Overview The input format of the system is a list of phonemes with assigned prosody descriptors similar to the MBROLA format [4], but extended by syllable and stress markers. A program to generate this extension automatically is provided by the system. The output are the parameters for the KLSYN88 synthesizer. For the experiments, the implementation of Sensymetrics Corp. was used. The data concept is hierarchical: An ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F. and Van der Vreken, O. The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for Non-Commercial Purposes, Proc ICSLP, 3:13931396, Philadelphia, 1996

....to selectively degrade or preserve the phonological and prosodic properties of speech [23] Basically, each sentence is resynthesised by feeding its phonetic transcription, the duration of each phoneme and the fundamental frequency curve into an adequate speech synthesiser. Here, we use mbrola [26] , which performs the synthesis through concatenation of diphones, using a French diphone database. Phonological properties of the synthesised sentences are manipulated by modifying the input data to the synthesis. Here we only use one of the four types of manipulation documented in [23] ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken, The mbrola project: Towards a set of high-quality speech synthesizers free of use for noncommercial purposes, in ICSLP'96, Philadelphia, 1996.

..... L utilisation d un systme incrmental est efficace et permet son amlioration priodique. On analyse les noncs non compris par la machine et l on augmente son vocabulaire en consquence, pour ainsi accrotre ses capacits de dialogue. Le systme HALPIN est capable de formuler une synthse vocale [2] de la rponse que l usager peut entendre si l ordinateur est quip de matriel audio. 4. CONCLUSIONS ET PERSPECTIVES Nous avons dvelopp une nouvelle mthodologie de recueil de corpus via le WWW pour obtenir rapidement des donnes originales, en vue de modliser les connaissances du domaine, les ....

DUTOIT, T., PAGEL, V., PIERRET, N., BATAILLE, F., VAN DER VRECKEN, O., The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for Non-Commercial Purposes, ICSLP'96, Philadelphia, vol. 3, pp. 1393-1396.

....completely neglected or is modeled concurrently with fundamental frequency. This is based on the assumption that the loudness contour is implied by the fundamental frequency of the utterance. Many concatenative synthesizers based on either linear prediction or overlap and add methods (e.g. MBROLA [13]) use the inherent loudness values in the diphone data and no other modeling is used. Although, in the case of MBROLA, the possibility to control loudness is currently being studied. Thus, prosody control is usually accomplished with three separate modules: prosodic boundary placement, segmental ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken. The MBROLA project: Towards a set of high quality speech synthesizers free of use for non commercial purposes. In Proc. ICSLP '96, volume 3, pages 1393--1396, Philadelphia, PA, October 1996.

....and recorded onto a DAT tape. The speech material was re digitised at 16 kHz, 16 bit during the transfer to a PC. A higher sampling rate was purposefully avoided in order to guarantee a sound quality similar to the one typically used in state of the art concatenative speech synthesis (e.g. [5]; see also Outlook below) 2.4. Pre selecfion A pre selection procedure was carried out in order to reduce the data to a subset of good quality. For this purpose, each speaker rated the other five speakers vocalisations. The 180 vocalisations (6 speakers 30 vocalisations speaker) were ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., & van der Vrecken, O. (1996). The MBROLA project: Towards a set of high quality speech synthesizers free of use for non commercial purposes, ICSLP 96, p. 1393-1396.

....a string of phoneme labels 5 as input, and aligns the labels with the signal, so as to provide the exact phoneme durations. The algorithm is an iterative procedure that optimises the acoustic match between the original speech sample and its synthetic equivalent produced by the algorithm Mbrola [6] (on the basis of a diphone database for each particular language) The usefulness of this type of segmentation to the issues of interest here has not been evaluated yet. 4.3.2. A rhythm extractor without segmentation Realising the difficulties encountered by segmentation algorithms, Galves et ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken. The mbrola project: Towards a set of high-quality speech synthesizers free of use for non-commercial purposes. In ICSLP'96, Philadelphia, 1996. MBROLA is freely available from http://tcts.fpms.ac.be/synthesis/mbrola.html.

....triggering their responses or (b) by an inappropriate rhythmical prosodic organization of these acoustic cues. 5.1. Targets We give here results of a last close shadowing experiment using synthetic stimuli produced by feeding two different concatenative sysnthesis systems one using MBROLA [7] and one using TDPSOLA [2] with the segmental durations and the appropriate stylization of the melody of the slow versions of experiment I. Seven copy synthesis targets were computed: 6 copy synthesis of target uttered by speakers j, g and p using MBROLA with the male voice fr1 referenced as ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., and van der Vrecken, O. The MBROLA project: towards a set of high quality speech synthesizers free of use for non commercial purposes. In Proceedings of the International Conference on Speech and Language Processing, volume 3, pages 1393--1396, Philadelphia - USA, 1996.

....present in the phonological structure. The interpreted structure is then converted to a parametric form depending on the signal generation method. The phonetic descriptions and timings can be used to select diphones and express their durations and pitch contours for output with the MBROLA system [1]. The phonetic details can also be used to augment copysynthesis parameters for the HLsyn quasi articulatory formant synthesizer [4] The timings and pitch information have even been used to manipulate the prosody of natural speech using PSOLA [2] 3. USE OF EXTENSIBLE MARK UP LANGUAGE The ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., Van der Vreken, O. (1996) "The MBROLA project: towards a set of high-quality speech synthesizers free of use for non-commercial purposes" Proc. ICSLP'96, Philadelphia, vol. 3, pp. 1393-1396

....and loads the driver into the virtual machine. The remaining components are implemented as clients to CTT Broker servers, Figure 2, example 3) including: a text to speech component using RULSYS (Carlson et al. 1982) for text to phone conversion plus GLOVE (Carlson et al. 1990) or MBROLA (Dutoit et al. 1996) synthesisers. Several Swedish and English voices are currently available, including Lukas (Filipson Bruce, 1997) and the Infovox 18 voices Ingmar, Annmarie and Roger. It can generate media streams for multi modal output (face and voice) Beskow, 1995) a StarLite speech recogniser (Strm, ....

Dutoit T, Pagel V, Pierret N, Batialle F & van der Vreken O (1996). The MBROLA project: towards a set of high-quality speech synthesizers free of use for non-commercial purposes. Proc of ICSLP, Philadelphia, 3:1393-1396.

....Architecture of the Media Centric system Client The real media client distributes the presentation streams to . the video rendering unit . avatar behaviour decision unit o Java 3D based feature morphing renderer, see [Alexa99] o phonem generator [Portele92] combined with a speech generator [Dutoit96] o viseme generator The avatar decision unit decides which avatar behaviour to show to the user. The avatar behaviour consists of three components: the behaviour streamed by the real server. Since this behaviour is streamed synchronous to the video (see figure 10) it is called synchronous ....

Dutoit, T. and Pagel, V. and Pierret, N. and Van Der Vreken O. and Bataille, F.: MBROLA Project: Towards a Set of HighQuality Speech Synthesizers Free of Use for Non-Commercial Purposes, Proceedings of the ICSLP 96, Philadelphia, 1996

....during enrolment, and the password phrase used during a test is generated by the system and given to the user through an audio prompt. 2.6. The text to speech system The text to speech (TTS) system consists of a rule based text to phone converter (Carlson et al. 1982) and an MBROLA synthesiser (Dutoit et al. 1996) with the Lukas voice (Filipson 5 The first difference to the baseline system in Melin Lindberg (1999) where selection was based on similarity to enrolment data rather than test data. 6 The second difference: in Melin Lindberg (1999) segmentation was created via forced alignment to what ....

Dutoit T, Pagel V, Pierret N, Batialle F & van der Vreken O (1996). The MBROLA project: towards a set of high-quality speech synthesizers free of use for non-commercial purposes. Proc of ICSLP, Philadelphia, 3:1393-1396.

....varies considerably among researchers, as it depends to a large extent on the criteria used in the unit selection processes and the design constraints imposed by the synthesis system. The most common approach is to define a set of synthesis units (e.g. diphones, triphones and demi syllables [4][5] and from this definition create a speech database that contains at least one example for each unit. Such an approach goes some way to overcoming some of the well known problems of coarticulation, where the production of one phone can be highly influenced by its preceding and following ....

Dutoit, T. et al, "The MBROLA project: towards a set of high quality speech synthesizers free of use for non commercial purposes", ICSLP' 96, 1996.

....with 3 These radicals were selected on the basis of a reanalysis of previous results (Dupoux, Kakehi, Hirose, Pallier, Mehler, in press) We chose from the consonants that had given the most robust effects in the past. CROSS LINGUISTIC PHONOLOGICAL PROCESSING 13 a MBROLA speech synthesizer (Dutoit, Pagel, Pierret, Bataille, Vreken, 1996), using the natural productions of a male Japanese speaker as a model. The algorithm for the speech synthesis used a male voice and a French diphone database. The test stimuli were edited with a speech editor to make sure there was no schwa vowel inserted in the consonant cluster; any evidence of ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., & Vreken, O. V. D. (1996). The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for Non-Commercial Purposes. Paper presented at the ICSLP'96, Philadelphia.

....by the TTS. The evaluation process consists of prosodic transplantation tasks where source and target signals are uttered by the same speaker. 2. Evaluation of AMSS: an ill posed problem Thanks to the emergence of statistical methods in speech synthesis, the freely available generic tools (Dutoit et al. 1996; Taylor et al. 1998) and multilingual resources, building a TTS is no longer necessarily the fruit of the strenuous work of speech scientists. It is thus very important to provide means to evaluate the intrinsic properties of each module of this giant generic lego . Few glass box evaluation ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., and van der Vrecken, O., 1996. The MBROLA project: towards a set of high quality speech synthesizers free of use for non commercial purposes. In Proceedings of the International Conference on Speech and Language Processing, volume 3. Philadelphia - USA.

....eyes to look at an approaching user [6] Speech synthesis parameter trajectories were generated by the KTH audio visual text to speech system. Apart from generating the appropriate lip movements in the animated face, these were also used as input to a Mbrola synthesizer for the sound generation [7]. The responses that were known in advance, including Strindberg quotations, were manually checked and changed. 5 CONCLUDING REMARKS So far, the August system has been used by about 3000 people, which has generated a database of spontaneous man machine interactions with the animated agent. The ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F. and van der Vreken, O. (1996) The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for Non-Commercial Purposes, In Proc. of ICSLP'96, Philadelphia, vol. 3, pp. 1393-1396

....allowing to selectively degrade or preserve the phonological and prosodic properties of speech. Basically, each sentence is resynthesized by feeding its phonetic transcription, the duration of each phoneme and the fundamental frequency curve into an adequate speech synthesizer. Here, we use MBROLA (Dutoit, Pagel, Pierret, Bataille, van der Vrecken, 1996) 6 , which performs the synthesis through concatenation of diphones, using a French diphone database. Phonological properties of the synthesized sentences are manipulated by modifying the input data to the synthesis. Here we only use two of the four types of manipulation documented in Ramus and ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., & van der Vrecken, O. (1996). The MBROLA Project: Towards a set of high-quality speech synthesizers free of use for non-commercial purposes. In ICSLP'96. Philadelphia.

....new communicative gestures in future dialogue systems. Acknowledgments The work with multimodal speech synthesis is done within the Centre for Speech Technology (CTT) at KTH. Some of the audio output of the text to speech synthesis used in the August system, is generated using an MBROLA database [18] for Swedish, cre ated by Marcus Filipsson and G sta Bruce of the Department of Linguistics and Phonetics of Lund University, Sweden. ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, O. van der Vreken, "The MBROLA Project: Towards a Set of HighQuality Speech Synthesizers Free of Use for NonCommercial Purposes" Proc. ICSLP'96, Philadelphia, vol. 3, pp. 1393-1396 .

....with image analysis software tools [14, 15] Speech synthesis parameter trajectories were generated by the KTH audio visual text to speech system. Apart from generating the appropriate lipmovements in the animated face, these were also used as input to a Mbrola synthesiser for the sound generation [16]. The responses that were known in advance, including Strindberg quotations, were manually labelled with prosodic information. to further enhance the phrasing. A common problem was the distribution of stress, phrasing and prominence [18] 5. EXTENDING THE SYSTEM One important topic, which has ....

Dutoit, T., Pagel, V., Pierret, N., Bataille, F. and van der Vreken, O. (1996) The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for NonCommercial Purposes, Proc. ICSLP'96, Philadelphia, vol. 3, pp. 1393-1396

....is provided with BREF, we generated a first segmentation to iterate the training process using a new boostrapping method [6] which is based on the use of a speech synthesizer for the alignment of a text on its corresponding speech signal. A high quality digital speech synthesizer (MBROLA [7]) was used to create a synthetic reference speech pattern which is aligned with the speech signal. So the alignment process is reduced to a simple dynamic time warping algorithm. In comparison to HMMs, this approach apparently looses speaker independence since only one voice is used as a ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille and O. Van Der Vreken, "The MBROLA Project : Toward a Set of High quality Speech Synthesizers Free of Use for non commercial purposes", International Conference on Speech and Language Processing, Philadelphia 1996.

.... is typically described in terms of fundamental frequency contours (for voiced portions of speech) and of duration of speech segments (mainly phonemes or syllables) Speech synthesizers driven with correct prosodic information are currently able to produce very high quality synthetic speech [1][2]. Different methods has been proposed for prosody generation, mainly rule based approaches [3] 4] or corpus based approaches based on automatic learning techniques like neural networks [5] 6] linear regression [7] or other statistical approaches [8] This communication presents a prosody ....

....of prosody: phonemic duration and f0 curve. The phonetic transcription and the syntactic structure of the text are computed with the morpho syntactic phonetizer LIPSS (Linguistic Processing for Speech Synthesis) t0] The synthetic prosody is given, with the sequence of phonemes, to the MBROLA [2] speech synthesizer to allow a qualitative evaluation of the prosody generated. Text . m J Symbolic Prosody [ Rhythm l r ] nerator [ Generationl I I MBROLA I I FoCurve I. 1 n [ Synthesizer I I Generation Synthetic L Speech Figure 1: Text to speech synthesizer. 2.1. Symbolic ....

Dutoit T., Pagel V., Pierret N., Bataille F. and van der Vrecken O., "The MBROLA Project: Towards a Set of High Quality Speech Synthesizers Free of Use for non commercial purposes" Proceedings of ICSLP96, pp. 1393-1396, 1996.

.... becomes possible to obtain more natural speech stimuli than those produced by methods such as random splicing [1] gated speech [2] rotated speech, reiterated speech, and many kinds of bandpass filterings [3] In this paper, we propose a new method for delexicalization based on the free MBROLA [4] speech synthesizer. We have performed a perception experiment based on this method, in order to elicit how listeners interpret F0 contours and segment duration variations in terms of prosodic boundaries, with a view to defining robust acoustic cues for the automatic segmentation of continuous ....

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken. The mbrola project: Towards a set of high-quality speech synthesizers free of use for non-commercial purpose. In ICSLP'96, Philadelphia, 1996.

.... is typically described in terms of fundamental frequency contours (for voiced portions of speech) and of duration of speech segments (mainly phonemes or syllables) Speech synthesizers driven with correct prosodic information are currently able to produce very high quality synthetic speech [1][2]. Various methods have been proposed for prosody generation, mainly rule based approaches [3] 4] or corpus based approaches based on automatic learning techniques like neural networks [5] 6] linear regression [7] or other statistical methods [8] This communication presents a prosody generation ....

.... durations and f 0 curve) In our TTS system, the phonetic transcription and the syntactic structure of the text are computed with the morpho syntactic phonetizer LIPSS (LInguistic Processing for Speech Synthesis) 10] The synthetic prosody is given, with the sequence of phonemes, to the MBROLA [2] speech synthesizer to allow a qualitative evaluation of the prosody generated. Figure 1 : Text to speech synthesizer. 2.1. Symbolic Description of Prosody The first task of the prosody generation system is to compute the prosodic phrasing of the input text. Grammatical tag Function word (G) ....

Dutoit T., Pagel V., Pierret N., Bataille F. and van der Vrecken O., " The MBROLA Project : Towards a Set of High Quality Speech Synthesizers Free of Use for non commercial purposes ", Proceedings of ICSLP'96, pp. 1393-1396, 1996.

....database is recursively refined using forced Viterbi alignment. As no phonetic segmentation is provided with BREF, we generated a first segmentation to iterate the training process using a new bootstrapping method [8] which is based on the use of a high quality digital speech synthesizer (MBROLA [4]) to create, from the phonetic transcription, a reference speech pattern with accurate phoneme boundaries and then align the natural speech on this pattern. So the alignment process is reduced to a simple dynamic time warping (DTW) algorithm (see fig 1) In comparison to HMMs, this approach ....

Dutoit T., Pagel V., Pierret N., Bataille F. and Van Der Vreken O., "The MBROLA Project : Toward a Set of High quality Speech Synthesizers Free of Use for non commercial purposes", Proceedings ICSLP'96, pp. 1393-1396, Philadelphia, PA, USA, 1996.

....the words that were immediately next to test entries are very similar. The question of accuracy with respect to genuinely unknown words is discussed more fully in [8] The automatic learning programs described in this paper as well as speaking dictionaries are available from the MBROLA project [9] home page http: tcts.fpms.ac.be synthesis mbrdico 6. ....

Dutoit T., Pagel V., Pierret N., Bataille F., Van der Vrecken O. "The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for NonCommercial Purpose", proc ICSLP'96,vol.3, p1393-1396

No context found.

Dutoit, Thierry, Vicent Pagel, N icolas Pierret, Francois Bataille and Olivier Van der Vrecken . 1996. The MBROLA Project: Towards a Set of H igh-Quality Speech Synthesizers Free of Use for Non Commercial Pu rpose . In Proceedings of ICSLP96,vol.3, p1393-1396.

No context found.

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., and der Vrecken, O. V. The MBROLA project: Towards a Set of High Quality Speech Synthesizers Free of Use for Non Commercial Purposes. In Proceedings of ICSLP '96 (Philadelphia, PA, October 1996), vol. 3, pp. 1393-1396. 148

No context found.

Dutoit, T., Pagel, V., Pierret, N., Bataille, F., Van Der Vreken, O., 1996. The MBROLA Project: towards a set of highquality speech synthesizers free of use for non-commercial purposes. In: Proc. ICSLP-96, Philadelphia, Vol. 3, pp. 1393-- 1396.

No context found.

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken, "The MBROLA project: Towards a set of high quality speech synthesisers free of use for non commercial purposes," in Proc. 4th ICSLP, Philadelphia, USA, 1996, pp. 1393--1396.

No context found.

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken, "The MBROLA project: Towards a set of high quality speech synthesisers free of use for non commercial purposes," in Proc. 4th ICSLP, Philadelphia, USA, 1996, pp. 1393--1396.

No context found.

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken. The MBROLA project: Towards a set of high quality speech synthesisers free of use for non commercial purposes. In Proc. 4th Int'l. Conf. of Spoken Language Processing, pages 1393--1396, 1996.

No context found.

T. Dutoit, V. Pagel, N. Pierret, O. van der Vreken, and F. Bataille. The mbrola project: Towards a set of high-quality speech synthesizers free of use for non-commercial purposes. In Proceedings of the Fourth International Conference on Spoken Language Processing, Philadelphia, USA, October 1996.

No context found.

T. Dutoit, V. Pagel, N. Pierret, F. Bataille, and O. van der Vrecken. The MBROLA project: Towards a set of high quality speech synthesisers free of use for non commercial purposes. In Proc. 4th Int'l. Conf. of Spoken Language Processing, pages 1393--1396, 1996.

No context found.

Dutoit, T., V, Pagel., N, Pierret., F, Bataille., O, Van Der Vrecken. (1996) "The MBROLA Project: Towards a Set of High-Quality Speech Synthesisers Free of Use for Non-Commercial Purposes". Proc ICSLP'96, Philadelphia, Vol. 3, pp. 1393-1396.

No context found.

Dutoit, T.; Pagel, V.; Pierret, N.; Bataille, F.; van der Vrecken, O., 1996. The MBROLA Project: Towards a Set of High-Quality Speech Synthesizers Free of Use for Non-Commercial Purposes. Proc. ICSLP 96, Philadelphia, vol. 3, pp. 1393-1396.

No context found.

T. Dutoit, V. Pagel, N. Pierret, O. van der Vreken, and F. Bataille, "The MBROLA project: Towards a set of high-quality speech synthesizers free of use for non-commercial purposes," in ICSLP96, Philadelphia, PA., 1996, vol. 3, pp. 1393--1397, http://tcts.fpms.ac.be/synthesis/mbrola.html.

No context found.

Dutoit, T., V. Pagel, N. Pierret, F. Bataille, and O. van der Vreken, (1996), "The MBROLA project: towards a set of highquality speech synthesizers free of use for non-commercial purposes, " Proc. 4 ICSLP '96, Philadelphia, vol.3, 13931396.

No context found.

Thierry Dutoit, Vincent Pagel, N. Pierret, F. Bataille, and O. van der Vrecken, "The MBROLA project: Towards a set of high quality speech synthesizers free for use for non commercial purposes," in Proc. International Conference on Spoken Language Processing (Philadelphia), 1996, vol. 3, pp. 1393--1396.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC