Update app.py

app.py

@@ -1,8 +1,8 @@
-
 import streamlit as st
 import openai
 import os
 from pydub import AudioSegment
+from pydub.silence import split_on_silence
 from dotenv import load_dotenv
 from tempfile import NamedTemporaryFile
 import math
@@ -14,54 +14,37 @@ load_dotenv()
 # Set your OpenAI API key
 openai.api_key = os.getenv("OPENAI_API_KEY")
 
-def get_chunk_length_ms(file_path, target_size_mb):
-    """
-    Calculate the length of each chunk in milliseconds to create chunks of approximately target_size_mb.
-
-    Args:
-    file_path (str): Path to the audio file.
-    target_size_mb (int): Target size of each chunk in megabytes.
-
-    Returns:
-    int: Chunk length in milliseconds.
-    """
-    audio = AudioSegment.from_file(file_path)
-    file_size_bytes = os.path.getsize(file_path)
-    duration_ms = len(audio)
-
-    # Calculate the approximate duration per byte
-    duration_per_byte = duration_ms / file_size_bytes
-
-    # Calculate the chunk length in milliseconds for the target size
-    chunk_length_ms = target_size_mb * 1024 * 1024 * duration_per_byte
-    return math.floor(chunk_length_ms)
-
-def split_audio(audio_file_path, chunk_length_ms):
+def split_audio_on_silence(audio_file_path, min_silence_len=500, silence_thresh=-40, keep_silence=250):
     """
-    Split an audio file into chunks of specified length.
-
+    Split an audio file into chunks using silence detection.
+    
     Args:
-    audio_file_path (str): Path to the audio file.
-    chunk_length_ms (int): Length of each chunk in milliseconds.
-
+        audio_file_path (str): Path to the audio file.
+        min_silence_len (int): Minimum length of silence (in ms) required to be used as a split point.
+        silence_thresh (int): The volume (in dBFS) below which is considered silence.
+        keep_silence (int): Amount of silence (in ms) to retain at the beginning and end of each chunk.
+    
     Returns:
-    list: List of AudioSegment chunks.
+        list: List of AudioSegment chunks.
     """
     audio = AudioSegment.from_file(audio_file_path)
-
-    chunks = [audio[i:i + chunk_length_ms] for i in range(0, len(audio), chunk_length_ms)]
-
+    chunks = split_on_silence(
+        audio, 
+        min_silence_len=min_silence_len,
+        silence_thresh=silence_thresh,
+        keep_silence=keep_silence
+    )
     return chunks
 
 def transcribe(audio_file):
     """
-    Transcribe an audio file using OpenAI Whisper model.
+    Transcribe an audio file using the OpenAI Whisper model.
 
     Args:
-    audio_file (str): Path to the audio file.
+        audio_file (str): Path to the audio file.
 
     Returns:
-    str: Transcribed text.
+        str: Transcribed text.
     """
     with open(audio_file, "rb") as audio:
         response = openai.audio.transcriptions.create(
@@ -77,10 +60,10 @@ def process_audio_chunks(audio_chunks):
     Process and transcribe each audio chunk.
 
     Args:
-    audio_chunks (list): List of AudioSegment chunks.
+        audio_chunks (list): List of AudioSegment chunks.
 
     Returns:
-    str: Combined transcription from all chunks.
+        str: Combined transcription from all chunks.
     """
     transcriptions = []
     min_length_ms = 100  # Minimum length required by OpenAI API (0.1 seconds)
@@ -107,11 +90,11 @@ def save_transcription_to_docx(transcription, audio_file_path):
     Save the transcription as a .docx file.
 
     Args:
-    transcription (str): Transcribed text.
-    audio_file_path (str): Path to the original audio file for naming purposes.
+        transcription (str): Transcribed text.
+        audio_file_path (str): Path to the original audio file for naming purposes.
 
     Returns:
-    str: Path to the saved .docx file.
+        str: Path to the saved .docx file.
     """
     # Extract the base name of the audio file (without extension)
     base_name = os.path.splitext(os.path.basename(audio_file_path))[0]
@@ -132,10 +115,9 @@ def save_transcription_to_docx(transcription, audio_file_path):
 
 st.title("Audio Transcription with OpenAI's Whisper")
 
-# uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "ogg", "m4a"])
+# Allow uploading of audio or video files
 uploaded_file = st.file_uploader("Upload an audio or video file", type=["wav", "mp3", "ogg", "m4a", "mp4", "mov"])
 
-
 if 'transcription' not in st.session_state:
     st.session_state.transcription = None
 
@@ -144,15 +126,19 @@ if uploaded_file is not None and st.session_state.transcription is None:
     
     # Save uploaded file temporarily
     file_extension = uploaded_file.name.split(".")[-1]
-    original_file_name = uploaded_file.name.rsplit('.', 1)[0]  # Get the original file name without extension
+    original_file_name = uploaded_file.name.rsplit('.', 1)[0]  # Get original file name without extension
     temp_audio_file = f"temp_audio_file.{file_extension}"
     with open(temp_audio_file, "wb") as f:
         f.write(uploaded_file.getbuffer())
 
-    # Split and process audio
+    # Split and process audio using silence detection
     with st.spinner('Transcribing...'):
-        chunk_length_ms = get_chunk_length_ms(temp_audio_file, target_size_mb=1)
-        audio_chunks = split_audio(temp_audio_file, chunk_length_ms)
+        audio_chunks = split_audio_on_silence(
+            temp_audio_file, 
+            min_silence_len=500,  # adjust based on your audio characteristics
+            silence_thresh=-40,   # adjust based on the ambient noise level
+            keep_silence=250      # optional: keeps a bit of silence at the edges
+        )
         transcription = process_audio_chunks(audio_chunks)
         if transcription:
             st.session_state.transcription = transcription
@@ -177,4 +163,3 @@ if st.session_state.transcription:
             file_name=st.session_state.output_docx_file,
             mime='application/vnd.openxmlformats-officedocument.wordprocessingml.document'
         )
-