Skip zero-length audio in embedding colab notebook.

PiperOrigin-RevId: 676521874

chirp/inference/embed_lib.py

@@ -261,6 +261,15 @@ def _log_exception(self, source_info, exception, counter_name):
         exception,
     )
 
+  def validate_audio(self, source_info: SourceInfo, audio: np.ndarray) -> bool:
+    if audio is None or audio.shape[0] == 0:
+      self._log_exception(source_info, 'no_exception', 'audio_empty')
+      return False
+    if audio.shape[0] < self.min_audio_s * self.target_sample_rate:
+      self._log_exception(source_info, 'no_exception', 'audio_too_short')
+      return False
+    return True
+
   def audio_to_example(
       self, file_id: str, timestamp_offset_s: float, audio: np.ndarray
   ) -> tf.train.Example:
@@ -345,11 +354,7 @@ def process(self, source_info: SourceInfo, crop_s: float = -1.0):
         self._log_exception(source_info, inst, 'audio_runtime_error')
       return
 
-    if audio is None or audio.shape[0] == 0:
-      self._log_exception(source_info, 'no_exception', 'audio_empty')
-      return
-    if audio.shape[0] < self.min_audio_s * self.target_sample_rate:
-      self._log_exception(source_info, 'no_exception', 'audio_too_short')
+    if not self.validate_audio(source_info, audio):
       return
 
     logging.info(

embed_audio.ipynb

@@ -202,6 +202,8 @@
     "      output_dir=output_dir, num_files=config.get('tf_record_shards', 1)) as file_writer:\n",
     "    for source_info, audio in tqdm.tqdm(\n",
     "        zip(new_source_infos, audio_iterator), total=len(new_source_infos)):\n",
+    "      if not embed_fn.validate_audio(source_info, audio):\n",
+    "        continue\n",
     "      file_id = source_info.file_id(config.embed_fn_config.file_id_depth)\n",
     "      offset_s = source_info.shard_num * source_info.shard_len_s\n",
     "      example = embed_fn.audio_to_example(file_id, offset_s, audio)\n",