ResumenPresenta una metodología novedosa para establecer una comunicación silenciosa dual basada en habla subvocal; para ello se desarrollaron dos sistemas electrónicos que registran las señales bioeléctricas que llegan al aparato fonador, generadas al momento de realizar el proceso de lectura silenciosa por el individuo. Estos sistemas están basados en tres etapas fundamentales; la primera es la de adquisición, encargada de extraer, acondicionar, codificar y transmitir las señales electromiográficas del habla subvocal hacia la segunda etapa, denominada de procesamiento; en esta etapa, implementada en un sistema Raspberry Pi, se desarrollaron los procesos de almacenamiento, acondicionamiento, extracción de patrones y clasificación de palabras, utilizando técnicas matemáticas como: Entropía, análisis Wavelet y Máquinas de Soporte Vectorial de Mínimos Cuadrados, implementadas bajo el entorno libre de programación Python; finalmente, la última etapa del sistema se encargó de comunicar inalámbricamente los dos sistemas electrónicos, utilizando 4 clases de señales, para clasificar las palabras hola, intruso, ¿hola cómo estás? y tengo frío.Adicionalmente, en este artículo se muestra la implementación del sistema para el registro de señales de habla subvocal. El porcentaje de acierto promedio general es de 72.5 %. Se incluyen un total de 50 palabras por clase, es decir, 200 señales. Finalmente, se pudo demostrar que usando una Raspberry Pi es posible establecer un sistema de comunicación silenciosa a partir de las señales del habla subvocal.Palabras clave: comunicación silenciosa; entropía; habla subvocal; MSV (Máquinas de Soporte Vectorial); Raspberry Pi; Wavelet. AbstractThis paper presents a novel methodology to develop a silent dual communication based on subvocal speech. Two electronic systems were developed for people's wireless communication. The system has 3 main stages. The first stage is the subvocal speech electromyographic signals acquisition, in charge to extract, condition, encode and transmit the system development. This signals were digitized and registered from the throat and sent to an embedded a raspberry pi.In this device was implemented the processing, as it is called the second stage, which besides to store, assumes conditioning, extraction and pattern classification of subvocal speech signals. Mathematical techniques were used as Entropy, Wavelet analysis, Minimal Squares and Vector Support Machines, which were applied in Python free environment program. Finally, in the last stage in charge to communicate by wireless means, were developed the two electronic systems, by using 4 signal types, to classify the words: Hello, intruder, hello how are you? and I am cold to perform the silent communication.Additionally, in this article we show the speech subvocal signals' recording system realization. The average accuracy percentage was 72.5 %, and includes a total of 50 words by class, this is 200 signals. Finally, it demonstrated that using the Raspberry Pi it is possible to set a silent communication ...

Subvocal electromyogram (EMG) signal classification is used to control a modified web browser interface. Recorded surface signals from the larynx and sublingual areas below the jaw are filtered and transformed into features using a complex dual quad tree wavelet transform. Feature sets for six subvocally pronounced control words, 10 digits, 17 vowel phonemes and 23 consonant phonemes are trained using a scaled conjugate gradient neural network. The subvocal signals are classified and used to initiate web browser queries through a matrix based alphabet coding scheme. Hyperlinks on web pages returned by the browser are numbered sequentially and queried using digits only. Classification methodology, accuracy, and feasibility for scale up to real world human machine interface tasks are discussed in the context of vowel and consonant recognition accuracy.

We present results of electromyographic (EMG) speech recognition on a small vocabulary of 15 English words. EMG speech recognition holds promise for mitigating the effects of high acoustic noise on speech intelligibility in communication systems, including those used by first responders (a focus of this work). We collected 150 examples per word of single-channel EMG data from a male subject, speaking normally while wearing a firefighter's self-contained breathing apparatus. The signal processing consisted of an activity detector, a feature extractor, and a neural network classifier. Testing produced an overall average correct classification rate on the 15 words of 74% with a 95% confidence interval of (71%, 77%). Once trained, the subject used a classifier as part of a real-time system to communicate to a cellular phone and to control a robotic device. These tasks were performed under an ambient noise level of approximately 95 decibels. We also describe ongoing work on phoneme-level EMG speech recognition. Crown