diff --git a/2_build_audio_vectors.ipynb b/2_build_audio_vectors.ipynb
index e8369d7..67cc38e 100644
--- a/2_build_audio_vectors.ipynb
+++ b/2_build_audio_vectors.ipynb
@@ -140,7 +140,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.5.2"
+   "version": "3.6.9"
   }
  },
  "nbformat": 4,
diff --git a/src/build_audio_vectors.py b/src/build_audio_vectors.py
new file mode 100644
index 0000000..9261910
--- /dev/null
+++ b/src/build_audio_vectors.py
@@ -0,0 +1,57 @@
+"""
+This script should be run AFTER extract_emotion_labels.py. It basically uses
+the csv file created in the previous step to split the original wav files into
+multiple smaller frames, each piece containing an emotion.
+
+Run this script from root as python src/build_audio_vectors.py
+"""
+
+import os
+import math
+import pickle
+import librosa
+import pandas as pd
+from tqdm import tqdm
+
+
+def process_session(iemocap_dir, labels_df, sr, sess):
+    """
+    saves audio_vectors dict in a pickle file which contains vectors
+    for audio files in session `sess`
+
+    process_session: Str pd.DataFrame Nat Int -> None
+    """
+    audio_vectors = {}
+    wav_file_path = '{}Session{}/dialog/wav/'.format(iemocap_dir, sess)
+    orig_wav_files = os.listdir(wav_file_path)
+    for orig_wav_file in tqdm(orig_wav_files):
+        try:
+            orig_wav_vector, _sr = librosa.load(wav_file_path + orig_wav_file,
+                                                sr=sr)
+            orig_wav_file, file_format = orig_wav_file.split('.')
+            for index, row in labels_df[labels_df['wav_file'].str.contains(
+                    orig_wav_file)].iterrows():
+                start_time, end_time, truncated_wav_file_name, = \
+                    row['start_time'], row['end_time'], row['wav_file']
+                start_frame = math.floor(start_time * sr)
+                end_frame = math.floor(end_time * sr)
+                truncated_wav_vector = orig_wav_vector[start_frame:end_frame+1]
+                audio_vectors[truncated_wav_file_name] = truncated_wav_vector
+        except Exception as e:
+            print('An exception occured for {}'.format(orig_wav_file))
+    with open('data/pre-processed/audio_vectors_{}.pkl'.format(sess), 'wb') as f:
+        pickle.dump(audio_vectors, f)
+
+
+def main():
+    sampling_rate = 44100
+    iemocap_dir = 'data/IEMOCAP_full_release/'
+    labels_df = pd.read_csv('data/pre-processed/df_iemocap.csv')
+    for sess in range(1, 6):
+        # Note that compiling this way will take too much time So you might
+        # consider parallelizing this process
+        process_session(iemocap_dir, labels_df, sampling_rate, sess)
+
+
+if __name__ == '__main__':
+    main()