[!88][DEV] Refactor to isolate libraries specific for speech translation

As people might need to work on projects not related to speech translation,
we want to avoid adding dependencies that are needed only for speech translation project
in the main setup file.
Conversely, we do not want to make difficult the installation for speech projects.

The patch creates a dedicated requirements txt file that contains the dependencies specific for ST projects and
moves the speech-only libraries to the new requirement file.

Existing CI

.github/workflows/build.yml

@@ -37,7 +37,7 @@ jobs:
         git submodule update --init --recursive
         python setup.py build_ext --inplace
         python -m pip install --editable .
-        pip install torchaudio
+        pip install -r speech_requirements.txt
         curdir=$(pwd) && cd ..
         git clone https://github.com/facebookresearch/SimulEval.git/
         cd SimulEval

.gitlab-ci.yml

@@ -19,7 +19,7 @@ before_script:
   - virtualenv venv
   - source venv/bin/activate
   - pip install -e .
-  - pip install torchaudio
+  - pip install -r speech_requirements.txt
   - curdir=$(pwd) && cd ..
   - rm -rf SimulEval
   - git clone https://github.com/facebookresearch/SimulEval.git/

FBK_HOW_TO_WORK.md

@@ -13,6 +13,8 @@ To start working on this code, first download the repository with `git clone`.
 The master branch containing the up-to-date FBK MT Fairseq internal version is `internal_master`,
 so you can access it entering into the cloned folder and running the command `git checkout internal_master`.
 To install the repository, run `pip install -e .`.
+If you plan to work on speech translation, complete the setup of you environment
+by installing the required dependencies with `pip install -r speech_requirements.txt`.
 We recommend installing the repository in a dedicated python virtual environment,
 which you can create with PyCharm when importing the project or on command line.
 In alternative, you can create a dedicated Anaconda environment.

README.md

@@ -25,6 +25,14 @@ Dedicated README for each work can be found in the `fbk_works` directory.
 If using this repository, please acknowledge the related paper(s) citing them.
 Bibtex citations are available for each work in the dedicated README file.
 
+To install the repository, do:
+
+```
+pip install -e .
+pip install -r speech_requirements.txt  # required for speech translation
+```
+
+
 Below, there is the original Fairseq README file.
 
 --------------------------------------------------------------------------------

examples/speech_to_text/scripts/from_srt_to_blocks.py

@@ -13,9 +13,9 @@
 # limitations under the License
 
 try:
-    import pysrt
+    import srt
 except ImportError:
-    print("Please install pysrt 'pip install pysrt'")
+    print("Please install srt 'pip install srt'")
     raise ImportError
 import re
 import sys
@@ -47,11 +47,12 @@ def main():
     and each newline inside that block will be substituted by an <eol>.
     """
     srt_path = sys.argv[1]
-    subs = pysrt.open(srt_path)
+    with open(srt_path) as f:
+        subs = list(srt.parse(f))
 
     with open(srt_path + ".blocks", 'w') as fp:
         for sub in subs:
-            fp.write("%s\n" % add_eol_eob(sub.text))
+            fp.write("%s\n" % add_eol_eob(sub.content))
 
 
 if __name__ == "__main__":

setup.py

@@ -192,8 +192,6 @@ def do_setup(package_data):
             "sacrebleu>=1.4.12",
             "torch",
             "tqdm",
-            "ctc_segmentation",
-            "pysrt"
         ],
         dependency_links=dependency_links,
         packages=find_packages(

speech_requirements.txt

@@ -0,0 +1,3 @@
+torchaudio
+ctc_segmentation
+srt