Con esta herramienta te facilitamos un acceso a todas las ofertas y demandas de tecnología europeas y a búsquedas de socios para participar en propuestas europeas de I+D publicadas en la red Enterprise Europe Network, pudiendo filtrar los resultados para facilitar las búsquedas más acordes con tus necesidades.

¿Quieres recibir estos listados de oportunidades de colaboración en tu correo de forma periódica y personalizada? Date de alta en nuestro Boletín

Para optimizar los resultados de la búsqueda, se recomienda utilizar términos en inglés.

Nuevo método para mejorar la síntesis de voz basado en algoritmo de selección de unidades

Resumen

Tipo:
Oferta Tecnológica
Referencia:
TOCZ20150629001
Publicado:
03/08/2015
Caducidad:
02/08/2016
Resumen:
Una universidad checa ha desarrollado y obtenido la patente de un nuevo método para mejorar la síntesis de voz basado en un algoritmo de selección de unidades. Este método evita la configuración manual de los sistemas de síntesis de voz. Los sistemas actuales a menudo se configuran manualmente. Este proceso manual generalmente incluye varias iteraciones. Con cada modificación o ajuste en el sistema, es necesario volver a sintetizar, escuchar y evaluar algunas frases de referencia. Este método evita estos pasos y permite detectar las deficiencias del sistema. La universidad busca empresas especializadas en desarrollo de software de síntesis de voz con el fin de establecer acuerdos de licencia.

Details

Tittle:
New method to tune speech synthesis systems based on unit-selection algorithm
Summary:
A Czech university has patented a new method to tune speech synthesis systems based on unit-selection algorithm. This method eliminates manual set up of speech synthesis systems. The university is looking for companies dealing with the development of speech synthesis software interested in license agreement.
Description:
Unit-selection speech synthesis is one of the most frequently used methods of computer speech synthesis. The principle of this method consists of an optimal selection of specific speech unit instances (phones, diphones, triphones or halfphones) and their smoothed concatenation. The selection of the optimal instances of speech segments is dynamically performed in synthesis runtime with respect to various phonetic, prosodic and acoustic specifications based on the synthesized utterance. The selection is usually controlled by two basic criteria called "costs" (target and concatenation costs) that are composed of subcomponents. At present, there is no universally accepted method dealing with proposal of components of individual costs and their weights. The components of the individual costs are usually designed on the basis of expert knowledge or experience obtained from experiments. The costs and their weights (including weights of cost subcomponents) are often tuned sub-optimally on the basis of limited listening tests or by using an objective method.

The Czech university has patented a new method to automatically tune the costs, subcomponents and their weights. This method is based on correlation between outlying (extreme) cost values and the corresponding listening-test-based annotation by human listeners. The listening test is performed only once for a given speech synthesis system (and its voice). The method consists of an analysis of an annotation or listening test of utterances synthesized by the tested synthesis system, detection of extremal values of speech-synthesis cost function and their combination. Results of the analysis can be used to automatically set up and train cost-function weights and also to analyze perceptual importance of individual weights.

The university is looking for companies focused on the development of speech synthesis software. Partner sought should be interested in implementation of the offered method to his own developed software under the license agreement.
Advantages and Innovations:
The current speech-synthesis systems are often set up manually. This manual process usually comprises several iterations. Every system modification or adjustment leads to a need to re-synthesize some reference sentences, to re-listen them and re-evaluate them. The offered method eliminates these steps and moreover enables to diagnose system weaknesses (i.e. origins of artefacts that degrade the quality of synthetic speech).
Stage of Development:
Field tested/evaluated
IPs:
Patents granted
CommeR Statunts Regarding IPR Status:
CZ patent

Partner sought

Type and Role of Partner Sought:
Type of the partner sought: Companies focused on the development of speech synthesis software.

Role of the partner sought: Implementation of a new method to tune a unit-selection speech synthesis system to his own developed software under the license agreement.

Client

Type and Size of Client:
University
Already Engaged in Trans-National Cooperation:
Si
Languages Spoken:
English

Keywords

Technology Keywords:
01003013 Information Technology/Informatics
01003017 Speech Processing/Technology