RESUMEN
Background: In today's world, numerous applications integral to various facets of daily life include automatic speech recognition methods. Thus, the development of a successful automatic speech recognition system can significantly augment the convenience of people's daily routines. While many automatic speech recognition systems have been established for widely spoken languages like English, there has been insufficient progress in developing such systems for less common languages such as Turkish. Moreover, due to its agglutinative structure, designing a speech recognition system for Turkish presents greater challenges compared to other language groups. Therefore, our study focused on proposing deep learning models for automatic speech recognition in Turkish, complemented by the integration of a language model. Methods: In our study, deep learning models were formulated by incorporating convolutional neural networks, gated recurrent units, long short-term memories, and transformer layers. The Zemberek library was employed to craft the language model to improve system performance. Furthermore, the Bayesian optimization method was applied to fine-tune the hyper-parameters of the deep learning models. To evaluate the model's performance, standard metrics widely used in automatic speech recognition systems, specifically word error rate and character error rate scores, were employed. Results: Upon reviewing the experimental results, it becomes evident that when optimal hyper-parameters are applied to models developed with various layers, the scores are as follows: Without the use of a language model, the Turkish Microphone Speech Corpus dataset yields scores of 22.2 -word error rate and 14.05-character error rate, while the Turkish Speech Corpus dataset results in scores of 11.5 -word error rate and 4.15 character error rate. Upon incorporating the language model, notable improvements were observed. Specifically, for the Turkish Microphone Speech Corpus dataset, the word error rate score decreased to 9.85, and the character error rate score lowered to 5.35. Similarly, the word error rate score improved to 8.4, and the character error rate score decreased to 2.7 for the Turkish Speech Corpus dataset. These results demonstrate that our model outperforms the studies found in the existing literature.
RESUMEN
Presentamos trabajo en progreso acerca de la normalización de palabras para contenidos generados por usuarios. El enfoque es simple y ayuda a reducir el volumen de anotaciones manuales características de enfoques más clásicos. Primero, agrupamos las variantes ortográficas de una palabra, mayormente las abreviaturas. De estos ejemplos agrupados manualmente aprendemos un clasificador automático que, dada una palabra no vista anteriormente, determina si es una variación ortográfica de una palabra conocida o si es una palabra totalmente nueva. Para lograr eso, calculamos la similitud entre la palabra no vista y todas las palabras conocidas, y clasificamos la nueva palabra como una variante ortográfica de su palabra más similar. El clasificador aplica una medida de similitud de secuencia de caracteres basada en la distancia de edición Levenshtein. Para mejorar la exactitud de esta medida, le asignamos a las operaciones de edición un costo basado en el error. Este esquema de asignación de costos apunta a maximizar la distancia entre secuencias similares que son variantes de diferentes palabras. Esta medida establecida de similitud alcanza una exactitud de .68, una importante mejoría si la comparamos con el .54 obtenido por la distancia Levenshtein.(AU)
We present work in progress on word normalization for user-generated content. The approach is simple and helps in reducing the amount of manual annotation characteristic of more classical approaches. First, orthographic variants of a word, mostly abbreviations, are grouped together. From these manually grouped examples,we learn an automated classifier that, given a previously unseen word, determines whether it is an orthographic variant of a known word or an entirely new word. To do that, we calculate the similarity between the unseen word and all known words, and classify the new word as an orthographic variant of its most similar word. The classifier applies a string similarity measure based on the Levenshtein edit distance. To improve the accuracy of this measure, we assign edit operations an error-based cost. This scheme of cost assigning aims to maximize the distance between similarstrings that are variants of different words. This custom similarity measure achieves an accuracy of .68, an important improvement if we compare it with the .54 obtained by the Levenshtein distance.(AU)
Asunto(s)
Psicología , Lenguaje , Habla , Reconocimiento de Normas Patrones AutomatizadasRESUMEN
Presentamos trabajo en progreso acerca de la normalización de palabras para contenidos generados por usuarios. El enfoque es simple y ayuda a reducir el volumen de anotaciones manuales características de enfoques más clásicos. Primero, agrupamos las variantes ortográficas de una palabra, mayormente las abreviaturas. De estos ejemplos agrupados manualmente aprendemos un clasificador automático que, dada una palabra no vista anteriormente, determina si es una variación ortográfica de una palabra conocida o si es una palabra totalmente nueva. Para lograr eso, calculamos la similitud entre la palabra no vista y todas las palabras conocidas, y clasificamos la nueva palabra como una variante ortográfica de su palabra más similar. El clasificador aplica una medida de similitud de secuencia de caracteres basada en la distancia de edición Levenshtein. Para mejorar la exactitud de esta medida, le asignamos a las operaciones de edición un costo basado en el error. Este esquema de asignación de costos apunta a maximizar la distancia entre secuencias similares que son variantes de diferentes palabras. Esta medida establecida de similitud alcanza una exactitud de .68, una importante mejoría si la comparamos con el .54 obtenido por la distancia Levenshtein.
We present work in progress on word normalization for user-generated content. The approach is simple and helps in reducing the amount of manual annotation characteristic of more classical approaches. First, orthographic variants of a word, mostly abbreviations, are grouped together. From these manually grouped examples,we learn an automated classifier that, given a previously unseen word, determines whether it is an orthographic variant of a known word or an entirely new word. To do that, we calculate the similarity between the unseen word and all known words, and classify the new word as an orthographic variant of its most similar word. The classifier applies a string similarity measure based on the Levenshtein edit distance. To improve the accuracy of this measure, we assign edit operations an error-based cost. This scheme of cost assigning aims to maximize the distance between similarstrings that are variants of different words. This custom similarity measure achieves an accuracy of .68, an important improvement if we compare it with the .54 obtained by the Levenshtein distance.