diff --git a/src/helm/benchmark/presentation/run_entries_speech.conf b/src/helm/benchmark/presentation/run_entries_audio.conf similarity index 100% rename from src/helm/benchmark/presentation/run_entries_speech.conf rename to src/helm/benchmark/presentation/run_entries_audio.conf diff --git a/src/helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py b/src/helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py index a9ab936e7c..cd8606fc1d 100644 --- a/src/helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py +++ b/src/helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py @@ -56,10 +56,14 @@ def get_instances(self, output_path: str) -> List[Instance]: if not file_name.endswith(".wav") or is_invalid_audio_file(local_audio_path): continue - answer: str = file_name.split("_")[-1].split(".")[0] input = Input( multimedia_content=MultimediaObject([MediaObject(content_type="audio/wav", location=local_audio_path)]) ) + + answer: str = file_name.split("_")[-1].split(".")[0] + if answer == "throatclearing": + answer = "throat clearing" + references = [Reference(Output(text=str(answer)), tags=[CORRECT_TAG])] instances.append(Instance(input=input, references=references, split=TEST_SPLIT)) return instances