Add initial project

.gitignore

@@ -0,0 +1,6 @@
+*.local
+.vscode
+.venv
+output
+input
+__pycache__

.streamlit/config.toml

@@ -0,0 +1,294 @@
+# Below are all the sections and options you can have in ~/.streamlit/config.toml.
+ 
+
+[global]
+
+# By default, Streamlit displays a warning when a user sets both a widget
+# default value in the function defining the widget and a widget value via
+# the widget's key in `st.session_state`.
+
+# If you'd like to turn off this warning, set this to True.
+
+# Default: false
+# disableWidgetStateDuplicationWarning = false
+
+# If True, will show a warning when you run a Streamlit-enabled script
+# via "python my_script.py".
+
+# Default: true
+# showWarningOnDirectExecution = true
+
+
+[logger]
+
+# Level of logging: 'error', 'warning', 'info', or 'debug'.
+
+# Default: 'info'
+# level = "info"
+
+# String format for logging messages. If logger.datetimeFormat is set,
+# logger messages will default to `%(asctime)s.%(msecs)03d %(message)s`. See
+# [Python's documentation](https://docs.python.org/2.6/library/logging.html#formatter-objects)
+# for available attributes.
+
+# Default: "%(asctime)s %(message)s"
+# messageFormat = "%(asctime)s %(message)s"
+
+
+[client]
+
+# Controls whether uncaught app exceptions and deprecation warnings
+# are displayed in the browser. By default, this is set to True and
+# Streamlit displays app exceptions and associated tracebacks, and
+# deprecation warnings, in the browser.
+
+# If set to False, deprecation warnings and full exception messages
+# will print to the console only. Exceptions will still display in the
+# browser with a generic error message. For now, the exception type and
+# traceback show in the browser also, but they will be removed in the
+# future.
+
+# Default: true
+# showErrorDetails = true
+
+# Change the visibility of items in the toolbar, options menu,
+# and settings dialog (top right of the app).
+
+# Allowed values:
+# * "auto" : Show the developer options if the app is accessed through
+# localhost or through Streamlit Community Cloud as a developer.
+# Hide them otherwise.
+# * "developer" : Show the developer options.
+# * "viewer" : Hide the developer options.
+# * "minimal" : Show only options set externally (e.g. through
+# Streamlit Community Cloud) or through st.set_page_config.
+# If there are no options left, hide the menu.
+
+# Default: "auto"
+# toolbarMode = "auto"
+
+# Controls whether the default sidebar page navigation in a multi-page app is displayed.
+
+# Default: true
+# showSidebarNavigation = true
+
+
+[runner]
+
+# Allows you to type a variable or string by itself in a single line of
+# Python code to write it to the app.
+
+# Default: true
+# magicEnabled = true
+
+# Handle script rerun requests immediately, rather than waiting for script
+# execution to reach a yield point. This makes Streamlit much more
+# responsive to user interaction, but it can lead to race conditions in
+# apps that mutate session_state data outside of explicit session_state
+# assignment statements.
+
+# Default: true
+# fastReruns = true
+
+# Raise an exception after adding unserializable data to Session State.
+# Some execution environments may require serializing all data in Session
+# State, so it may be useful to detect incompatibility during development,
+# or when the execution environment will stop supporting it in the future.
+
+# Default: false
+# enforceSerializableSessionState = false
+
+# Adjust how certain 'options' widgets like radio, selectbox, and
+# multiselect coerce Enum members when the Enum class gets
+# re-defined during a script re-run.
+
+# Allowed values:
+# * "off": Disables Enum coercion.
+# * "nameOnly": Enum classes can be coerced if their member names match.
+# * "nameAndValue": Enum classes can be coerced if their member names AND
+# member values match.
+
+# Default: "nameOnly"
+# enumCoercion = "nameOnly"
+
+
+[server]
+
+# List of folders that should not be watched for changes. This
+# impacts both "Run on Save" and @st.cache.
+
+# Relative paths will be taken as relative to the current working directory.
+
+# Example: ['/home/user1/env', 'relative/path/to/folder']
+
+# Default: []
+# folderWatchBlacklist = []
+
+# Change the type of file watcher used by Streamlit, or turn it off
+# completely.
+
+# Allowed values:
+# * "auto" : Streamlit will attempt to use the watchdog module, and
+# falls back to polling if watchdog is not available.
+# * "watchdog" : Force Streamlit to use the watchdog module.
+# * "poll" : Force Streamlit to always use polling.
+# * "none" : Streamlit will not watch files.
+
+# Default: "auto"
+# fileWatcherType = "auto"
+
+# Symmetric key used to produce signed cookies. If deploying on multiple replicas, this should
+# be set to the same value across all replicas to ensure they all share the same secret.
+
+# Default: randomly generated secret key.
+# cookieSecret = "75ce94a30d4b5d3c91b99ec42cca35d49077b6c620d9c20187e2ed8a64bf3053"
+
+# If false, will attempt to open a browser window on start.
+
+# Default: false unless (1) we are on a Linux box where DISPLAY is unset, or
+# (2) we are running in the Streamlit Atom plugin.
+# headless = false
+
+# Automatically rerun script when the file is modified on disk.
+
+# Default: false
+# runOnSave = false
+
+# The address where the server will listen for client and browser
+# connections. Use this if you want to bind the server to a specific address.
+# If set, the server will only be accessible from this address, and not from
+# any aliases (like localhost).
+
+# Default: (unset)
+# address =
+
+# The port where the server will listen for browser connections.
+
+# Default: 8501
+# port = 8501
+
+# The base path for the URL where Streamlit should be served from.
+
+# Default: ""
+# baseUrlPath = ""
+
+# Enables support for Cross-Origin Resource Sharing (CORS) protection, for added security.
+
+# Due to conflicts between CORS and XSRF, if `server.enableXsrfProtection` is on and
+# `server.enableCORS` is off at the same time, we will prioritize `server.enableXsrfProtection`.
+
+# Default: true
+# enableCORS = true
+
+# Enables support for Cross-Site Request Forgery (XSRF) protection, for added security.
+
+# Due to conflicts between CORS and XSRF, if `server.enableXsrfProtection` is on and
+# `server.enableCORS` is off at the same time, we will prioritize `server.enableXsrfProtection`.
+
+# Default: true
+# enableXsrfProtection = true
+
+# Max size, in megabytes, for files uploaded with the file_uploader.
+
+# Default: 200
+maxUploadSize = 2048
+
+# Max size, in megabytes, of messages that can be sent via the WebSocket connection.
+
+# Default: 200
+maxMessageSize = 2048
+
+# Enables support for websocket compression.
+
+# Default: false
+# enableWebsocketCompression = false
+
+# Enable serving files from a `static` directory in the running app's directory.
+
+# Default: false
+# enableStaticServing = false
+
+# Server certificate file for connecting via HTTPS.
+# Must be set at the same time as "server.sslKeyFile".
+
+# ['DO NOT USE THIS OPTION IN A PRODUCTION ENVIRONMENT. It has not gone through security audits or performance tests. For the production environment, we recommend performing SSL termination by the load balancer or the reverse proxy.']
+# sslCertFile =
+
+# Cryptographic key file for connecting via HTTPS.
+# Must be set at the same time as "server.sslCertFile".
+
+# ['DO NOT USE THIS OPTION IN A PRODUCTION ENVIRONMENT. It has not gone through security audits or performance tests. For the production environment, we recommend performing SSL termination by the load balancer or the reverse proxy.']
+# sslKeyFile =
+
+
+[browser]
+
+# Internet address where users should point their browsers in order to
+# connect to the app. Can be IP address or DNS name and path.
+
+# This is used to:
+# - Set the correct URL for CORS and XSRF protection purposes.
+# - Show the URL on the terminal
+# - Open the browser
+
+# Default: "localhost"
+# serverAddress = "localhost"
+
+# Whether to send usage statistics to Streamlit.
+
+# Default: true
+# gatherUsageStats = true
+
+# Port where users should point their browsers in order to connect to the
+# app.
+
+# This is used to:
+# - Set the correct URL for CORS and XSRF protection purposes.
+# - Show the URL on the terminal
+# - Open the browser
+
+# Default: whatever value is set in server.port.
+# serverPort = 8501
+
+
+[mapbox]
+
+# Configure Streamlit to use a custom Mapbox
+# token for elements like st.pydeck_chart and st.map.
+# To get a token for yourself, create an account at
+# https://mapbox.com. It's free (for moderate usage levels)!
+
+# Default: ""
+# token = ""
+
+
+[deprecation]
+
+# Set to false to disable the deprecation warning for using the global pyplot instance.
+
+# Default: true
+# showPyplotGlobalUse = true
+
+
+[theme]
+
+# The preset Streamlit theme that your custom theme inherits from.
+# One of "light" or "dark".
+# base =
+
+# Primary accent color for interactive elements.
+# primaryColor =
+
+# Background color for the main content area.
+# backgroundColor =
+
+# Background color used for the sidebar and most interactive widgets.
+# secondaryBackgroundColor =
+
+# Color used for almost all text.
+# textColor =
+
+# Font family for all text in the app, except code blocks. One of "sans serif",
+# "serif", or "monospace".
+# font =
+

Home.py

@@ -0,0 +1,64 @@
+import streamlit as st
+import models.file as mf
+import models.video as mv
+import models.whisper as mw
+import models.subtitles as ms
+import models.transcript as mt
+
+def sidebar():
+    device = st.sidebar.selectbox('Select Device',('CPU','GPU'))
+    st.sidebar.write('You selected:', device)
+    return device
+
+def load_model():
+    # if pipeline is in session state, return it
+    # else, load it and save it to session state
+    if 'pipeline' not in st.session_state:
+        device = mw.get_device()
+        st.session_state['pipeline'] = mw.get_pipe(device)
+    return st.session_state['pipeline']
+
+def app():
+    device = sidebar()
+    pipeline = load_model()
+    st.title('Transcript Small Whisper')
+    st.write('Welcome to the Home page!')
+
+    file = st.file_uploader("Upload Files",type=['mp4','wav','mp3'])
+
+    progress_bar = st.progress(0)
+    status_text = st.empty()
+
+    if file is not None:
+        status_text.text('Uploading file...')
+        progress_bar.progress(10)
+        st.write(file.name)
+        status_text.text('File uploaded!')
+        file_details = {"FileName":file.name,"FileType":file.type,"FileSize":file.size}
+        st.write(file_details)
+        # st.write("type of file: ", type(file))
+        # st.write("Dir: ", dir(file))
+        # st.write("File: ", file.read())
+        if mf.get_file_type(file) == 'video':
+            status_text.text('Extracting audio from video...')
+            audio = mv.get_audio_from_video(file, file.name + '.mp3')
+        if mf.get_file_type(file) == 'audio':
+            status_text.text('Extracting audio from audio...')
+            audio = file.read()
+        progress_bar.progress(30)
+        status_text.text('Transcribing audio...')
+        transcript = mw.get_prediction(pipeline, audio)
+        progress_bar.progress(60)
+        status_text.text('Subtitling audio...')
+        subtitles = mw.get_prediction_with_timelines(transcript, file.name + '.srt')
+        progress_bar.progress(90)
+        status_text.text('Saving transcript...')
+        ms.save_subtitles(subtitles, file.name + '.srt')
+        status_text.text('Saving subtitles...')
+        mt.save_transcript(transcript, file.name + '.txt')
+        status_text.text('Done!')
+        progress_bar.progress(100)
+
+
+if __name__ == '__main__':
+    app()

README.md

@@ -5,8 +5,12 @@ colorFrom: blue
 colorTo: blue
 sdk: streamlit
 sdk_version: 1.30.0
-app_file: app.py
+app_file: Home.py
 pinned: false
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+# Transcript Small Whisper
+
+Creates a Transcript and a Subtitle from Video or Audio

models/audio.py

@@ -0,0 +1,16 @@
+from pydub import AudioSegment
+import os
+import math
+
+def split_audio(input_file, output_folder, duration):
+    audio = AudioSegment.from_mp3(input_file)
+    total_length = len(audio)
+    num_parts = math.ceil(total_length / (duration * 1000))
+ 
+    for i in range(num_parts):
+        start = i * duration * 1000
+        end = (i + 1) * duration * 1000
+        split_audio = audio[start:end]
+        output_path = os.path.join(output_folder, f"part_{i+1}.mp3")
+        split_audio.export(output_path, format="mp3")
+        print(f"Exported {output_path}")

models/file.py

@@ -0,0 +1,8 @@
+def get_file_type(file):
+    if 'video' in file.type:
+        return 'video'
+    elif 'audio' in file.type:
+        return 'audio'
+    else:
+        return None
+        

models/subtitles.py

@@ -0,0 +1,9 @@
+
+def save_subtitles(prediction, output_file):
+    ## Save prediction to subtitles file format
+    with open(output_file, "w") as f:
+        for i, chunk in enumerate(prediction):
+            f.write(f"{i+1}")
+            f.write(f"{chunk['timestamp'][0]} --> {chunk['timestamp'][1]}")
+            f.write(f"{chunk['text']}\n")
+            f.write("\n")

models/transcript.py

@@ -0,0 +1,5 @@
+
+def save_transcript(prediction, output_file):
+    ## Save prediction to subtitles file format
+    with open(output_file, "w") as f:
+        f.write(prediction)

models/video.py

@@ -0,0 +1,6 @@
+from moviepy.editor import *
+
+def get_audio_from_video(input_file, output_folder):
+    video = VideoFileClip(input_file)
+    video.audio.write_audiofile(output_folder)
+    print(f"Exported {output_folder}")

models/whisper.py

@@ -0,0 +1,58 @@
+import torch_directml
+from transformers import pipeline
+
+MODEL_CHECKPOINT = "openai/whisper-small"
+CHUNK_LENGTH_S = 30
+
+def get_device():
+    # return "cpu"
+    return torch_directml.device()
+
+def get_pipe(device, model_checkpoint=MODEL_CHECKPOINT, chunk_length_s=CHUNK_LENGTH_S):
+    return pipeline(
+        "automatic-speech-recognition",
+        model=model_checkpoint,
+        chunk_length_s=chunk_length_s,
+        device=device,
+    )
+
+def get_prediction_with_timelines(pipe, sample):
+    return pipe(sample, batch_size=8, return_timestamps=True)["chunks"]
+
+def get_prediction(pipe, sample):
+    return pipe(sample, batch_size=8)["text"]
+
+
+# audio_file = open("movie.mp3", "rb")
+# sample = audio_file.read()
+# prediction = pipe(sample, batch_size=8, return_timestamps=True)["chunks"]
+
+
+# def main():
+#     print("# Extracting audio from video")
+#     # audio = get_audio_from_video("movie.mp4", "movie.mp3")
+#     print("# Splitting audio into chunks")
+#     split_audio("movie.mp3", "movie_parts", 30)
+#     print("# Gettin Device")
+#     device = get_device()
+#     print("# Getting Pipeline")
+#     pipe = get_pipe(device)
+#     print("# Getting Predictions")
+#     for file in os.listdir("movie_parts"):
+#         print(f"Processing {file}")
+#         audio_file = open(os.path.join("movie_parts", file), "rb")
+#         sample = audio_file.read()
+#         print(f"# Getting Prediction for file {file}")
+#         prediction = get_prediction(pipe, sample)
+#         print(f"# Saving Subtitles for file {file}")
+#         save_subtitles(prediction, file.replace(".mp3", ".srt"))
+#     # Aggregating subtitles into one file
+#     print("# Aggregating subtitles into one file")
+#     with open("movie.srt", "w") as f:
+#         for file in os.listdir("movie_parts"):
+#             with open(os.path.join("movie_parts", file), "r") as f2:
+#                 f.write(f2.read())
+#     print("# Done")
+
+# if __name__ == "__main__":
+#     main()

requirements.txt

@@ -0,0 +1,5 @@
+streamlit
+moviepy
+pydub
+torch
+transformers