Relates #31 | Add SEACrowd Speech Classification

seacrowd/utils/constants.py

@@ -14,6 +14,8 @@
     ssp_features,
     speech_text_features,
     speech2speech_features,
+    speech_features,
+    speech_multi_features,
     image_text_features,
 )
 
@@ -78,6 +80,11 @@ class Tasks(Enum):
     # SpeechText
     SPEECH_RECOGNITION = "ASR"
     SPEECH_TO_TEXT_TRANSLATION = "STTT"
+
+    SPEECH_LANGUAGE_IDENTIFICATION = "SPEECH_LID"
+    SPEECH_EMOTION_RECOGNITION = "SER"
+    SPEECH_EMOTION_RECOGNITION_MULTILABEL = "SER_MULTI"
+
     TEXT_TO_SPEECH = "TTS"
 
     # SpeechSpeech
@@ -205,6 +212,9 @@ class Licenses(Enum):
     Tasks.SPEECH_TO_TEXT_TRANSLATION: "SPTEXT",
     Tasks.TEXT_TO_SPEECH: "SPTEXT",
     Tasks.SPEECH_TO_SPEECH_TRANSLATION: "S2S",
+    Tasks.SPEECH_LANGUAGE_IDENTIFICATION: "SP_CLS",
+    Tasks.SPEECH_EMOTION_RECOGNITION: "SP_CLS",
+    Tasks.SPEECH_EMOTION_RECOGNITION_MULTILABEL: "SP_CLS_MULTI",
     Tasks.IMAGE_CAPTIONING: "IMTEXT",
     Tasks.STYLIZED_IMAGE_CAPTIONING: "IMTEXT",
     Tasks.VISUALLY_GROUNDED_REASONING: "IMTEXT",
@@ -234,6 +244,8 @@ class Licenses(Enum):
     "SSP": ssp_features,
     "SPTEXT": speech_text_features,
     "S2S": speech2speech_features,
+    "SP_CLS": speech_features(),
+    "SP_CLS_MULTI": speech_multi_features(),
     "IMTEXT": image_text_features(),
 }
 

seacrowd/utils/schemas/__init__.py

@@ -10,6 +10,8 @@
 from .self_supervised_pretraining import features as ssp_features
 from .speech_text import features as speech_text_features
 from .speech_to_speech import features as speech2speech_features
+from .speech_classification import features as speech_features
+from .speech_multilabel import features as speech_multi_features
 from .image_text import features as image_text_features
 
-__all__ = ["kb_features", "qa_features", "text2text_features", "text_features", "text_multi_features", "pairs_features", "pairs_multi_features", "pairs_features_score", "seq_label_features", "ssp_features", "speech_text_features", "speech2speech_features", "image_text_features"]
+__all__ = ["kb_features", "qa_features", "text2text_features", "text_features", "text_multi_features", "pairs_features", "pairs_multi_features", "pairs_features_score", "seq_label_features", "ssp_features", "speech_text_features", "speech2speech_features", "speech_features", "speech_multi_features", "image_text_features"]

seacrowd/utils/schemas/speech_classification.py

@@ -0,0 +1,19 @@
+"""
+Speech Classification Schema for Single Label (be it Binary or Multiclass)
+"""
+import datasets
+
+def features(label_names = ["Yes", "No"]):
+    return datasets.Features(
+        {
+            "id": datasets.Value("string"),
+            "path": datasets.Value("string"),
+            "audio": datasets.Audio(sampling_rate=16_000),
+            "speaker_id": datasets.Value("string"),
+            "labels": datasets.ClassLabel(names=label_names),
+            "metadata": {
+                "speaker_age": datasets.Value("int64"),
+                "speaker_gender": datasets.Value("string"),
+            }
+        }
+    )

seacrowd/utils/schemas/speech_multilabel.py

@@ -0,0 +1,19 @@
+"""
+Speech Classification Schema for Multilabel
+"""
+import datasets
+
+def features(label_names = ["Yes", "No"]):
+    return datasets.Features(
+        {
+            "id": datasets.Value("string"),
+            "path": datasets.Value("string"),
+            "audio": datasets.Audio(sampling_rate=16_000),
+            "speaker_id": datasets.Value("string"),
+            "labels": datasets.Sequence(datasets.ClassLabel(names=label_names)),
+            "metadata": {
+                "speaker_age": datasets.Value("int64"),
+                "speaker_gender": datasets.Value("string"),
+            }
+        }
+    )