Add TTS bundle/pipelines (#1872)

Future work items:
- length computation of GriffinLim
- better way to make InverseMelScale work in inference_mode

Author: mthrok

docs/source/pipelines.rst

@@ -167,6 +167,69 @@ HUBERT_ASR_XLARGE
 .. container:: py attribute
 
    .. autodata:: HUBERT_ASR_XLARGE
+
+
+Tacotron2 Text-To-Speech
+------------------------
+
+Tacotron2TTSBundle
+~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: Tacotron2TTSBundle
+
+   .. automethod:: get_text_processor
+
+   .. automethod:: get_tacotron2
+
+   .. automethod:: get_vocoder
+
+Tacotron2TTSBundle - TextProcessor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: torchaudio.pipelines::Tacotron2TTSBundle.TextProcessor
+   :members: tokens
+   :special-members: __call__
+
+
+Tacotron2TTSBundle - Vocoder
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: torchaudio.pipelines::Tacotron2TTSBundle.Vocoder
+   :members: sample_rate
+   :special-members: __call__
+
+
+TACOTRON2_WAVERNN_PHONE_LJSPEECH
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. container:: py attribute
+
+   .. autodata:: TACOTRON2_WAVERNN_PHONE_LJSPEECH
+      :no-value:
+
+
+TACOTRON2_WAVERNN_CHAR_LJSPEECH
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. container:: py attribute
+
+   .. autodata:: TACOTRON2_WAVERNN_CHAR_LJSPEECH
+      :no-value:
+
+TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. container:: py attribute
+
+   .. autodata:: TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH
+      :no-value:
+
+TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. container:: py attribute
+
+   .. autodata:: TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH
       :no-value:
 
 References

docs/source/refs.bib

@@ -1,3 +1,9 @@
+@misc{ljspeech17,
+  author       = {Keith Ito and Linda Johnson},
+  title        = {The LJ Speech Dataset},
+  howpublished = {\url{https://keithito.com/LJ-Speech-Dataset/}},
+  year         = {2017}
+}
 @misc{conneau2020unsupervised,
       title={Unsupervised Cross-lingual Representation Learning for Speech Recognition}, 
       author={Alexis Conneau and Alexei Baevski and Ronan Collobert and Abdelrahman Mohamed and Michael Auli},

test/integration_tests/tacotron2_pipeline_test.py

@@ -0,0 +1,28 @@
+from torchaudio.pipelines import (
+    TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH,
+    TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH,
+    TACOTRON2_WAVERNN_CHAR_LJSPEECH,
+    TACOTRON2_WAVERNN_PHONE_LJSPEECH,
+)
+import pytest
+
+
+@pytest.mark.parametrize(
+    'bundle',
+    [
+        TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH,
+        TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH,
+        TACOTRON2_WAVERNN_CHAR_LJSPEECH,
+        TACOTRON2_WAVERNN_PHONE_LJSPEECH,
+    ]
+)
+def test_tts_models(bundle):
+    """Smoke test of TTS pipeline"""
+    text = "Hello world! Text to Speech!"
+
+    processor = bundle.get_text_processor()
+    tacotron2 = bundle.get_tacotron2()
+    vocoder = bundle.get_vocoder()
+    processed, lengths = processor(text)
+    mel_spec, lengths, _ = tacotron2.infer(processed, lengths)
+    waveforms, lengths = vocoder(mel_spec, lengths)

torchaudio/pipelines/__init__.py

@@ -20,6 +20,13 @@
     HUBERT_ASR_LARGE,
     HUBERT_ASR_XLARGE,
 )
+from ._tts import (
+    Tacotron2TTSBundle,
+    TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH,
+    TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH,
+    TACOTRON2_WAVERNN_CHAR_LJSPEECH,
+    TACOTRON2_WAVERNN_PHONE_LJSPEECH,
+)
 
 __all__ = [
     'Wav2Vec2Bundle',
@@ -42,4 +49,9 @@
     'HUBERT_XLARGE',
     'HUBERT_ASR_LARGE',
     'HUBERT_ASR_XLARGE',
+    'Tacotron2TTSBundle',
+    'TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH',
+    'TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH',
+    'TACOTRON2_WAVERNN_CHAR_LJSPEECH',
+    'TACOTRON2_WAVERNN_PHONE_LJSPEECH',
 ]

torchaudio/pipelines/_tts/__init__.py

@@ -0,0 +1,16 @@
+from .interface import Tacotron2TTSBundle
+from .impl import (
+    TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH,
+    TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH,
+    TACOTRON2_WAVERNN_CHAR_LJSPEECH,
+    TACOTRON2_WAVERNN_PHONE_LJSPEECH,
+)
+
+
+__all__ = [
+    'Tacotron2TTSBundle',
+    'TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH',
+    'TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH',
+    'TACOTRON2_WAVERNN_CHAR_LJSPEECH',
+    'TACOTRON2_WAVERNN_PHONE_LJSPEECH',
+]