Code

app.py

@@ -3,6 +3,7 @@ from huggingface_hub import hf_hub_download
 from audio_index import AudioEmbeddingSystem
 from search import search
 import pandas as pd
+import numpy as np
 
 db_file = hf_hub_download(
     repo_id="freddyaboulton/common-voice-english-audio",
@@ -21,6 +22,8 @@ audio_embedding_system = AudioEmbeddingSystem(db_path=db_file, index_path=index_
 def audio_search(audio_tuple):
     sample_rate, array = audio_tuple
     array = array[: int(sample_rate * 10)]
+    array = array.astype(np.float32) / 32768.0
+
     rows = audio_embedding_system.search((sample_rate, array))
     orig_rows = search(rows)
     for row in rows: