Add audio support to DataPack

.github/workflows/main.yml

@@ -78,7 +78,7 @@ jobs:
  rm -rf texar-pytorch
  - name: Install Forte
  run: |
- pip install --use-feature=in-tree-build --progress-bar off .[ner,test,example,wikipedia,augment,stave,audio]
+ pip install --use-feature=in-tree-build --progress-bar off .[ner,test,example,wikipedia,augment,stave,audio_ext]
  - name: Build ontology
  run: |
  ./scripts/build_ontology_specs.sh

docs/index.rst

@@ -11,6 +11,7 @@ Welcome to Forte's documentation!
 
  examples.md
  ontology_generation.md
+ audio_processing.md
 
 API
 ====

forte/data/readers/__init__.py

@@ -30,3 +30,4 @@
 from forte.data.readers.ag_news_reader import *
 from forte.data.readers.largemovie_reader import *
 from forte.data.readers.misc_readers import *
+from forte.data.readers.audio_reader import *

forte/data/readers/audio_reader.py

@@ -16,7 +16,6 @@
 """
 import os
 from typing import Any, Iterator
-import soundfile
 
 from forte.data.data_pack import DataPack
 from forte.data.data_utils_io import dataset_path_iterator
@@ -30,6 +29,21 @@
 class AudioReader(PackReader):
  r""":class:`AudioReader` is designed to read in audio files."""
 
+ def initialize(self, resources, configs):
+ # pylint: disable=attribute-defined-outside-init
+ super().initialize(resources, configs)
+ try:
+ import soundfile # pylint: disable=import-outside-toplevel
+ except ModuleNotFoundError as e:
+ raise ModuleNotFoundError(
+ "AudioReader requires 'soundfile' package to be installed."
+ " You can run 'pip install soundfile' or 'pip install forte"
+ "[audio_ext]'. Note that additional steps might apply to Linux"
+ " users (refer to "
+ "https://pysoundfile.readthedocs.io/en/latest/#installation)."
+ ) from e
+ self._soundfile = soundfile
+
  def _collect(self, audio_directory) -> Iterator[Any]: # type: ignore
  r"""Should be called with param ``audio_directory`` which is a path to a
  folder containing audio files.
@@ -48,7 +62,7 @@ def _parse_pack(self, file_path: str) -> Iterator[DataPack]:
  pack: DataPack = DataPack()
 
  # Read in audio data and store in DataPack
- audio, sample_rate = soundfile.read(
+ audio, sample_rate = self._soundfile.read(
  file=file_path, **(self.configs.read_kwargs or {})
  )
  pack.set_audio(audio=audio, sample_rate=sample_rate)

setup.py

@@ -50,7 +50,7 @@
  # transformers 4.10.0 will break the translation model we used here
  "augment": ["transformers>=3.1, <=4.9.2", "nltk"],
  "stave": ["stave>=0.0.1.dev12"],
- "audio": ["soundfile>=0.10.3"],
+ "audio_ext": ["soundfile>=0.10.3"],
  },
  entry_points={
  "console_scripts": [

tests/forte/data/readers/audio_reader_test.py

@@ -24,7 +24,7 @@
 from forte.common.resources import Resources
 from forte.common.exception import ProcessFlowException
 from forte.data.data_pack import DataPack
-from forte.data.readers.audio_reader import AudioReader
+from forte.data.readers import AudioReader
 from forte.pipeline import Pipeline
 from forte.processors.base.pack_processor import PackProcessor