Merge pull request #39 from yash-fn/master

Dry-run + ability to specify models + minor edits

Author: abhirooptalasila

README.md

@@ -77,8 +77,8 @@ In the age of OTT platforms, there are still some who prefer to download movies/
     * For a GPU build that is reusable (saving time on instantiating the program):
     ```bash
     $ docker build --build-arg BASEIMAGE=nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04 --build-arg DEPSLIST=requirements-gpu.txt -t autosub-base . && \
-    docker run --gpus all --name autosub-base autosub-base || \
-    docker commit autosub-base autosub-instance
+    docker run --gpus all --name autosub-base autosub-base --dry-run || \
+    docker commit --change 'CMD []' autosub-base autosub-instance
     ```
     * Then
     ```bash

autosub/audioProcessing.py

@@ -4,7 +4,7 @@
 import sys
 import subprocess
 import numpy as np
-
+from os.path import basename
 
 def extract_audio(input_file, audio_file_name):
     """Extract audio from input video file and save to audio/ in root dir
@@ -17,8 +17,8 @@ def extract_audio(input_file, audio_file_name):
     try:
         command = ["ffmpeg", "-hide_banner", "-loglevel", "warning", "-i", input_file, "-ac", "1", "-ar", "16000",
                    "-vn", "-f", "wav", audio_file_name]
-        ret = subprocess.call(command)
-        print("Extracted audio to audio/{}".format(audio_file_name.split("/")[-1]))
+        ret = subprocess.run(command).returncode
+        print("Extracted audio to audio/{}".format(basename(audio_file_name)))
     except Exception as e:
         print("Error: ", str(e))
         sys.exit(1)

autosub/main.py

@@ -107,52 +107,130 @@ def ds_process_audio(ds, audio_file, output_file_handle_dict, split_duration):
 def main():
     global line_count
     print("AutoSub\n")
-
-    for x in os.listdir():
-        if x.endswith(".pbmm"):
-            print("Model: ", os.path.join(os.getcwd(), x))
-            ds_model = os.path.join(os.getcwd(), x)
-        if x.endswith(".scorer"):
-            print("Scorer: ", os.path.join(os.getcwd(), x))
-            ds_scorer = os.path.join(os.getcwd(), x)
-
-    # Load DeepSpeech model
-    try:
-        ds = Model(ds_model)
-    except:
-        print("Invalid model file. Exiting\n")
-        sys.exit(1)
-
-    try:
-        ds.enableExternalScorer(ds_scorer)
-    except:
-        print("Invalid scorer file. Running inference using only model file\n")
-
+    
+    #Parsing Arguments Section
     supported_output_formats = ['srt', 'vtt', 'txt']
     parser = argparse.ArgumentParser(description="AutoSub")
-    parser.add_argument('--file', required=True,
-                        help='Input video file')
+    
+    #Runtime Options
     parser.add_argument('--format', choices=supported_output_formats, nargs='+',
                         help='Create only certain output formats rather than all formats',
                         default=supported_output_formats)
     parser.add_argument('--split-duration', type=float,
                         help='Split run-on sentences exceededing this duration (in seconds) into multiple subtitles',
                         default=5)
-    args = parser.parse_args()
+    parser.add_argument('--dry-run', dest="dry_run", action="store_true",
+                        help="Perform dry-run to verify options prior to running. Also useful to instantiate cuda/tensorflow cache prior to running multiple times.")
 
-    if os.path.isfile(args.file):
-        input_file = args.file
-        print("\nInput file:", input_file)
+    #Files that should be supplied
+    #File no longer required here, but will check manually later
+    #Basically EITHER file OR dry-run is sufficient
+    parser.add_argument('--file', required=False, 
+                        help='Input video file')
+    parser.add_argument('--model', required=False,
+                        help='Input *.pbmm model file')
+    parser.add_argument('--scorer', required=False,
+                        help='Input *.scorer file')
+    
+    args = parser.parse_args()
+    
+    #Please keep the following because I need it for verifying dockerfiles.
+    print(sys.argv[0:])
+    print("ARGS:", args)
+
+    def getmodel(args, arg_name):
+        #prioritize supplied argument
+        
+        if arg_name == 'model':
+            arg_extension = '.pbmm'
+        elif arg_name == 'scorer':
+            arg_extension = '.scorer'
+        else:
+            print("Coding Error. This function only accepts model or scorer for arg_name.")
+            sys.exit(1)
+
+        arg = args.__getattribute__(arg_name)
+        
+        if arg is not None:
+            model = os.path.abspath(arg)
+            if not os.path.isfile(model):
+                print(f"Error. Supplied file {arg} doesn't exist. Please supply a valid {arg_name} file via the --{arg_name} flag.")
+                sys.exit(1)
+        else:
+        #try to find local models
+            models_ = os.listdir()
+            models = []
+            for file in models_:
+                if file.endswith(arg_extension):
+                    models.append(file) 
+            del(models_)
+
+            num_models = len(models)
+        
+            if num_models == 0:
+                print(f"Warning no {arg_name}s specified via --{arg_name} and none found in local directory. Please run getmodel.sh convenience script from autosub repo to get some.")
+                if arg_name == 'model':
+                    print("Error: Must have pbmm model. Exiting")
+                    sys.exit(1)
+                else:
+                    model = ''
+            elif num_models != 1: 
+                print(f"Warning. Detected multiple *{arg_extension} files in local dir. You must specify which one you wish to use via the --{arg_name} field. Details: \n {num_models} {models}")
+                if arg_name == 'model':
+                    print("Error: Must specify pbmm model. Exiting")
+                    sys.exit(1)
+                else:
+                    print("Since I cannot know which scorer you wish to use, I just won't use any and try to run inference without it.")
+                    model = ''
+            else:
+                model = os.path.abspath(models[0])                    
+        
+        print(f"{arg_name}: ", model)
+        return(model)
+ 
+    def InstantiateModel(model, scorer):
+        # Load DeepSpeech model
+        try:
+            ds = Model(model)
+        except:
+            print("Invalid model file. Exiting\n")
+            sys.exit(1)
+
+        try:
+            ds.enableExternalScorer(scorer)
+        except:
+            print("Invalid scorer file. Running inference using only model file\n")
+        return(ds)
+    
+
+    ds_model = getmodel(args, 'model')
+    ds_scorer = getmodel(args, 'scorer')
+
+    if args.dry_run:
+        InstantiateModel(ds_model, ds_scorer) 
+        if args.file is not None:
+            if not os.path.isfile(args.file):
+                print(f"Error: {args.file}: No such file exists")
+        sys.exit(0)
+
+    #Not a dry-run
+    if args.file is not None:
+        if os.path.isfile(args.file):
+            input_file = args.file
+            print("\nInput file:", input_file)
+        else:
+            print(args.file, ": No such file exists")
+            sys.exit(1)
     else:
-        print(args.file, ": No such file exists")
+        print("Error. You must supply a file with --file or to instantiate cuda cache you must supply a --dry-run.")
         sys.exit(1)
 
     base_directory = os.getcwd()
     output_directory = os.path.join(base_directory, "output")
     audio_directory = os.path.join(base_directory, "audio")
     video_prefix = os.path.splitext(os.path.basename(input_file))[0]
     audio_file_name = os.path.join(audio_directory, video_prefix + ".wav")
-
+ 
     output_file_handle_dict = {}
     for format in args.format:
         output_filename = os.path.join(output_directory, video_prefix + "." + format)
@@ -183,11 +261,25 @@ def main():
 
     print("\nRunning inference:")
 
-    for filename in tqdm(sort_alphanumeric(os.listdir(audio_directory))):
-        # Only run inference on relevant files, and don't run inference on the original audio file
-        if filename.startswith(video_prefix) and (filename != os.path.basename(audio_file_name)):
-            audio_segment_path = os.path.join(audio_directory, filename)
-            ds_process_audio(ds, audio_segment_path, output_file_handle_dict, split_duration=args.split_duration)
+    #Remove master audio file
+    audiofiles=sort_alphanumeric(os.listdir(audio_directory))
+    audiofiles.remove(os.path.basename(audio_file_name))
+
+    #Remove non related audiofiles potentially from other instances of autosub
+    audiofiles_ = []
+    for filename in audiofiles:
+        if filename.startswith(video_prefix):
+            audiofiles_.append(filename)
+    audiofiles = audiofiles_
+    del(audiofiles_)
+
+    #Process Segments
+
+    ds = InstantiateModel(ds_model, ds_scorer) 
+
+    for filename in tqdm(audiofiles):
+        audio_segment_path = os.path.join(audio_directory, filename)
+        ds_process_audio(ds, audio_segment_path, output_file_handle_dict, split_duration=args.split_duration)
 
     print("\n")
     for format in output_file_handle_dict:

scripts/gpu-build.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
 cd "$(dirname "$(readlink -f "$0")")"
 cd ..
-docker build --build-arg BASEIMAGE=nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04 --build-arg DEPSLIST=requirements-gpu.txt -t autosub-base . && docker run --gpus all --name autosub-base autosub-base || docker commit autosub-base autosub-instance && docker container rm autosub-base
+docker build --build-arg BASEIMAGE=nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04 --build-arg DEPSLIST=requirements-gpu.txt -t autosub-base . && docker run --gpus all --name autosub-base autosub-base --dry-run && docker commit --change 'CMD []' autosub-base autosub-instance && docker container rm autosub-base