Make TTS automodels importable (#25595)

* Add auto model for spectrogram/waveform

* Add doc and install

* Add dummy objects

* Did I miss anything?

docs/source/en/model_doc/auto.md

@@ -330,6 +330,14 @@ The following auto classes are available for the following audio tasks.
 
 [[autodoc]] AutoModelForAudioXVector
 
+### AutoModelForTextToSpectrogram
+
+[[autodoc]] AutoModelForTextToSpectrogram
+
+### AutoModelForTextToWaveform
+
+[[autodoc]] AutoModelForTextToWaveform
+
 ## Multimodal
 
 The following auto classes are available for the following multimodal tasks.

src/transformers/__init__.py

@@ -1135,6 +1135,8 @@ else:
             "AutoModelForSpeechSeq2Seq",
             "AutoModelForTableQuestionAnswering",
             "AutoModelForTextEncoding",
+            "AutoModelForTextToSpectrogram",
+            "AutoModelForTextToWaveform",
             "AutoModelForTokenClassification",
             "AutoModelForUniversalSegmentation",
             "AutoModelForVideoClassification",
@@ -5050,6 +5052,8 @@ if TYPE_CHECKING:
             AutoModelForSpeechSeq2Seq,
             AutoModelForTableQuestionAnswering,
             AutoModelForTextEncoding,
+            AutoModelForTextToSpectrogram,
+            AutoModelForTextToWaveform,
             AutoModelForTokenClassification,
             AutoModelForUniversalSegmentation,
             AutoModelForVideoClassification,

src/transformers/models/auto/__init__.py

@@ -101,6 +101,8 @@ else:
         "AutoModelForSequenceClassification",
         "AutoModelForSpeechSeq2Seq",
         "AutoModelForTableQuestionAnswering",
+        "AutoModelForTextToSpectrogram",
+        "AutoModelForTextToWaveform",
         "AutoModelForTokenClassification",
         "AutoModelForUniversalSegmentation",
         "AutoModelForVideoClassification",
@@ -280,6 +282,8 @@ if TYPE_CHECKING:
             AutoModelForSpeechSeq2Seq,
             AutoModelForTableQuestionAnswering,
             AutoModelForTextEncoding,
+            AutoModelForTextToSpectrogram,
+            AutoModelForTextToWaveform,
             AutoModelForTokenClassification,
             AutoModelForUniversalSegmentation,
             AutoModelForVideoClassification,

src/transformers/utils/dummy_pt_objects.py

@@ -742,6 +742,20 @@ class AutoModelForTextEncoding(metaclass=DummyObject):
         requires_backends(self, ["torch"])
 
 
+class AutoModelForTextToSpectrogram(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class AutoModelForTextToWaveform(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 class AutoModelForTokenClassification(metaclass=DummyObject):
     _backends = ["torch"]
 

utils/update_metadata.py

@@ -115,6 +115,7 @@ PIPELINE_TAGS_AND_AUTO_MODELS = [
     ("depth-estimation", "MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES", "AutoModelForDepthEstimation"),
     ("video-classification", "MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING_NAMES", "AutoModelForVideoClassification"),
     ("mask-generation", "MODEL_FOR_MASK_GENERATION_MAPPING_NAMES", "AutoModelForMaskGeneration"),
+    ("text-to-audio", "MODEL_FOR_TEXT_TO_SPECTROGRAM_NAMES", "AutoModelForTextToSpectrogram"),
     ("text-to-audio", "MODEL_FOR_TEXT_TO_WAVEFORM_NAMES", "AutoModelForTextToWaveform"),
 ]