Make ASR pipeline compliant with Hub spec + add tests (#33769)

* Remove max_new_tokens arg

* Add ASR pipeline to testing

* make fixup

* Factor the output test out into a util

* Full error reporting

* Full error reporting

* Update src/transformers/pipelines/automatic_speech_recognition.py

Co-authored-by: Lysandre Debut <hi@lysand.re>

* Small comment

---------

Co-authored-by: Lysandre Debut <hi@lysand.re>

tests/pipelines/test_pipelines_audio_classification.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 
 import unittest
-from dataclasses import fields
 
 import numpy as np
 from huggingface_hub import AudioClassificationOutputElement
@@ -21,6 +20,7 @@ from huggingface_hub import AudioClassificationOutputElement
 from transformers import MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING, TF_MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING
 from transformers.pipelines import AudioClassificationPipeline, pipeline
 from transformers.testing_utils import (
+    compare_pipeline_output_to_hub_spec,
     is_pipeline_test,
     nested_simplify,
     require_tf,
@@ -68,10 +68,8 @@ class AudioClassificationPipelineTests(unittest.TestCase):
 
         self.run_torchaudio(audio_classifier)
 
-        spec_output_keys = {field.name for field in fields(AudioClassificationOutputElement)}
         for single_output in output:
-            output_keys = set(single_output.keys())
-            self.assertEqual(spec_output_keys, output_keys, msg="Pipeline output keys do not match HF Hub spec!")
+            compare_pipeline_output_to_hub_spec(single_output, AudioClassificationOutputElement)
 
     @require_torchaudio
     def run_torchaudio(self, audio_classifier):