Update README.md

README.md

@@ -182,6 +182,24 @@ The dataset was collected as part of two research projects, CLIMAD (Cross-lingui
 
 Data were recorded between 2021 and 2024 using Shure SM-35 unidirectional cardioid microphones and Marantz PMD620 recorders, ensuring minimal noise interference. Recordings were captured at 48 kHz, 16-bit resolution [TO BE CONFIRMED]. Some of the recordings were annotated with orthographic and/or phonetic transcriptions and aligned at a word and phoneme level. Metadata includes speaker characteristics, language status (L1, L2, L3/Ln), task type, and audio details. 
 
+#### Data Preprocessing
+
+At first an audio file is normalized to a single channel, 16 kHz sample rate, and 2-byte sample width.   
+
+The recorded speech is analyzed in 30-second chunks. For such a preset time slice, the silence level is defined as 0.1 of the maximal loudness relative to the minimum loudness.  
+
+With preset values of the minimum silence length set, the recording above the silence level is treated as speech. 2 conditions are taken into account:  
+
+- The minimum length of recorded speech must not be shorter than 3 sec and not longer than 30 seconds 
+- In the case of a fragment shorter than 3 seconds, it is merged with the previously detected one  
+
+In addition, 30-second fragments are analyzed for the presence of silence considering the context of the entire recording. 
+
+The beginning of the next 30-second chunk is defined as the moment when the previous chunk ended. 
+
+
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/64a3dc4593bbb85b68b94ecb/qOZ0Fym5xXLGdunqFopKd.png)
+
 #### Who are the source data producers?
 
 <!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->