Better task descriptions

app.py

@@ -26,7 +26,7 @@ def inference(wav,data):
       speech, rate = soundfile.read(wav)
       if len(speech.shape)==2:
           speech=speech[:,0] 
-      if data == "english_slurp":
+      if data == "English intent classification and named entity recognition task based on the SLURP database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -52,7 +52,7 @@ def inference(wav,data):
             slot_val=k.split(" FILL ")[1]
             text=text+" "+slot_name+" : "+slot_val+","
           text=text+"}"
-      elif data == "english_fsc":
+      elif data == "English intent classification task based on the FSC database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -71,7 +71,7 @@ def inference(wav,data):
           objects=intent.split("_")[1]
           location=intent.split("_")[2]
           text="INTENT: {action: "+action+", object: "+objects+", location: "+location+"}"
-      elif data == "english_snips":
+      elif data == "English intent classification task based on the SNIPS database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -87,7 +87,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("in:","")
           text="INTENT: "+intent
-      elif data == "dutch_scr":
+      elif data == "Dutch speech command recognition task based on the Grabo database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -103,7 +103,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0]
           text="SPEECH COMMAND: "+intent
-      elif data == "english_scr":
+      elif data == "English speech command recognition task based on the Google Speech Commands database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -119,7 +119,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("command:","")
           text="SPEECH COMMAND: "+intent
-      elif data == "lithuanian_scr":
+      elif data == "Lithuanian speech command recognition task based on the Lithuanian SC database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -135,7 +135,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text
           text="SPEECH COMMAND: "+intent
-      elif data == "arabic_scr":
+      elif data == "Arabic speech command recognition task based on the Arabic SC database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -151,7 +151,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("command:","")
           text="SPEECH COMMAND: "+intent
-      elif data == "lid_voxforge":
+      elif data == "Language Identification task based on the VoxForge database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -166,7 +166,7 @@ def inference(wav,data):
         #   import pdb;pdb.set_trace()
           lang=speech2text.converter.tokenizer.tokenizer.convert_ids_to_tokens(nbests[0][2][0]).replace("|>","").replace("<|","")
           text="LANG: "+lang
-      elif data == "fake_speech_detection_asvspoof":
+      elif data == "English Fake Speech Detection task based on the ASVSpoof database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -182,7 +182,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("class:","")
           text="SPEECH CLASS: "+intent
-      elif data == "emotion_rec_iemocap":
+      elif data == "English emotion recognition task based on the IEMOCAP database":
           replace_dict={}
           replace_dict["em:neu"]="Neutral"
           replace_dict["em:ang"]="Angry"
@@ -203,7 +203,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=replace_dict[text.split(" ")[0]]
           text="EMOTION: "+intent
-      elif data == "accent_classify_accentdb":
+      elif data == "English accent classification task based on the Accent DB database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -219,7 +219,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("accent:","")
           text="ACCENT: "+intent
-      elif data == "sarcasm_mustard":
+      elif data == "English sarcasm detection task based on the MUStARD database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -235,7 +235,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("class:","")
           text="SARCASM CLASS: "+intent
-      elif data == "sarcasm_mustard_plus":
+      elif data == "English sarcasm detection task based on the MUStARD++ database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -251,7 +251,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("class:","")
           text="SARCASM CLASS: "+intent
-      elif data == "gender_voxceleb1":
+      elif data == "English gender identification task based on the VoxCeleb1 database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -267,7 +267,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("gender:f","female").replace("gender:m","male")
           text="GENDER: "+intent
-      elif data == "audio_classification_esc50":
+      elif data == "Audio classification task based on the ESC-50 database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -283,7 +283,7 @@ def inference(wav,data):
           text=text.split("|>")[-1]
           intent=text.split(" ")[0].replace("audio_class:","")
           text="AUDIO EVENT CLASS: "+audio_class_arr[int(intent)]
-      elif data == "semantic_parsing_stop":
+      elif data == "English semantic parsing task based on the STOP database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -299,7 +299,7 @@ def inference(wav,data):
           text, *_ = nbests[0]
           text=text.split("|>")[-1].replace("_STOP","")
           text="SEMANTIC PARSE SEQUENCE: "+text
-      elif data == "vad_freesound":
+      elif data == "Voice activity detection task based on the Google Speech Commands v2 and Freesound database":
           speech2text = Speech2Text.from_pretrained(
             asr_train_config="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/config.yaml",
             asr_model_file="UniverSLU-17-Task-Specifier/exp/asr_train_asr_whisper_full_correct_specaug2_copy_raw_en_whisper_multilingual/valid.acc.ave_10best.pth",
@@ -328,12 +328,12 @@ title = "UniverSLU"
 description = "Gradio demo for UniverSLU Task Specifier (https://huggingface.co/espnet/UniverSLU-17-Task-Specifier). UniverSLU-17 Task Specifier is a Multi-task Spoken Language Understanding model from CMU WAVLab. It adapts Whisper to additional tasks using single-token task specifiers. To use it, simply record your audio or click one of the examples to load them. More details about the SLU tasks that the model is trained on and it's performance on these tasks can be found in our paper: https://aclanthology.org/2024.naacl-long.151/"
 article = "<p style='text-align: center'><a href='https://github.com/espnet/espnet' target='_blank'>Github Repo</a></p>"
 
-examples=[['audio_slurp_ner.flac',"english_slurp"],['audio_fsc.wav',"english_fsc"],['audio_grabo.wav',"dutch_scr"],['audio_english_scr.wav',"english_scr"],['audio_lt_scr.wav',"lithuanian_scr"],['audio_ar_scr.wav',"arabic_scr"],['audio_snips.wav',"english_snips"],['audio_lid.wav',"lid_voxforge"],['audio_fsd.wav',"fake_speech_detection_asvspoof"],['audio_er.wav',"emotion_rec_iemocap"],['audio_acc.wav',"accent_classify_accentdb"],['audio_mustard.wav',"sarcasm_mustard"],['audio_mustard_plus.wav',"sarcasm_mustard_plus"],['audio_voxceleb1.wav',"gender_voxceleb1"],['audio_esc50.wav',"audio_classification_esc50"],['audio_stop.wav',"semantic_parsing_stop"],['audio_freesound.wav',"vad_freesound"]]
+examples=[['audio_slurp_ner.flac',"English intent classification and named entity recognition task based on the SLURP database"],['audio_fsc.wav',"English intent classification task based on the FSC database"],['audio_grabo.wav',"Dutch speech command recognition task based on the Grabo database"],['audio_english_scr.wav',"English speech command recognition task based on the Google Speech Commands database"],['audio_lt_scr.wav',"Lithuanian speech command recognition task based on the Lithuanian SC database"],['audio_ar_scr.wav',"Arabic speech command recognition task based on the Arabic SC database"],['audio_snips.wav',"English intent classification task based on the SNIPS database"],['audio_lid.wav',"Language Identification task based on the VoxForge database"],['audio_fsd.wav',"English Fake Speech Detection task based on the ASVSpoof database"],['audio_er.wav',"English emotion recognition task based on the IEMOCAP database"],['audio_acc.wav',"English accent classification task based on the Accent DB database"],['audio_mustard.wav',"English sarcasm detection task based on the MUStARD database"],['audio_mustard_plus.wav',"English sarcasm detection task based on the MUStARD++ database"],['audio_voxceleb1.wav',"English gender identification task based on the VoxCeleb1 database"],['audio_esc50.wav',"Audio classification task based on the ESC-50 database"],['audio_stop.wav',"English semantic parsing task based on the STOP database"],['audio_freesound.wav',"Voice activity detection task based on the Google Speech Commands v2 and Freesound database"]]
 
 # gr.inputs.Textbox(label="input text",lines=10),gr.inputs.Radio(choices=["english"], type="value", default="english", label="language")
 gr.Interface(
     inference, 
-    [gr.Audio(label="input audio",sources=["microphone"],type="filepath"),gr.Radio(choices=["english_slurp","english_fsc","dutch_scr","english_scr","lithuanian_scr","arabic_scr","english_snips","lid_voxforge","fake_speech_detection_asvspoof","emotion_rec_iemocap","accent_classify_accentdb","sarcasm_mustard","sarcasm_mustard_plus","gender_voxceleb1","audio_classification_esc50","semantic_parsing_stop","vad_freesound"], type="value", label="Task")], 
+    [gr.Audio(label="input audio",sources=["microphone"],type="filepath"),gr.Radio(choices=["English intent classification and named entity recognition task based on the SLURP database","English intent classification task based on the FSC database","Dutch speech command recognition task based on the Grabo database","English speech command recognition task based on the Google Speech Commands database","Lithuanian speech command recognition task based on the Lithuanian SC database","Arabic speech command recognition task based on the Arabic SC database","English intent classification task based on the SNIPS database","Language Identification task based on the VoxForge database","English Fake Speech Detection task based on the ASVSpoof database","English emotion recognition task based on the IEMOCAP database","English accent classification task based on the Accent DB database","English sarcasm detection task based on the MUStARD database","English sarcasm detection task based on the MUStARD++ database","English gender identification task based on the VoxCeleb1 database","Audio classification task based on the ESC-50 database","English semantic parsing task based on the STOP database","Voice activity detection task based on the Google Speech Commands v2 and Freesound database"], type="value", label="Task")], 
     gr.Textbox(type="text", label="Output"),
     title=title,
     cache_examples=False,