cal commited on
Commit ·
2da24ca
1
Parent(s): 287506c
speaches_audio_transcriber
Browse files
app.py
ADDED
|
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import os
|
| 3 |
+
import tempfile
|
| 4 |
+
from openai import OpenAI # Import the OpenAI client
|
| 5 |
+
import pandas as pd # Import pandas for data manipulation and Excel export
|
| 6 |
+
|
| 7 |
+
# Ensure your OpenAI API key is set as an environment variable.
|
| 8 |
+
# Example (for testing, but prefer environment variable):
|
| 9 |
+
# os.environ["OPENAI_API_KEY"] = "YOUR_ACTUAL_OPENAI_API_KEY_HERE"
|
| 10 |
+
# Make sure to replace "YOUR_ACTUAL_OPENAI_API_KEY_HERE" with your key
|
| 11 |
+
# if you choose to hardcode it (not recommended for production).
|
| 12 |
+
|
| 13 |
+
# Define the maximum file size for OpenAI's Whisper API (25 MB)
|
| 14 |
+
# Keeping this variable for reference, though chunking logic is removed
|
| 15 |
+
MAX_FILE_SIZE_MB = 25
|
| 16 |
+
MAX_FILE_SIZE_BYTES = MAX_FILE_SIZE_MB * 1024 * 1024
|
| 17 |
+
|
| 18 |
+
def transcribe_verbose(client, audio_file_path):
|
| 19 |
+
"""
|
| 20 |
+
Transcribes an audio file using OpenAI's Whisper model,
|
| 21 |
+
requesting verbose JSON output for segment details.
|
| 22 |
+
This function is now used directly for the full file.
|
| 23 |
+
"""
|
| 24 |
+
with open(audio_file_path, "rb") as audio_file:
|
| 25 |
+
transcription = client.audio.transcriptions.create(
|
| 26 |
+
model="whisper-1",
|
| 27 |
+
file=audio_file,
|
| 28 |
+
response_format="verbose_json" # Request verbose JSON to get segment times
|
| 29 |
+
)
|
| 30 |
+
return transcription
|
| 31 |
+
|
| 32 |
+
def format_time(seconds):
|
| 33 |
+
"""Converts seconds to HH:MM:SS.ms format."""
|
| 34 |
+
minutes, seconds = divmod(seconds, 60)
|
| 35 |
+
hours, minutes = divmod(minutes, 60)
|
| 36 |
+
return f"{int(hours):02}:{int(minutes):02}:{seconds:06.3f}"
|
| 37 |
+
|
| 38 |
+
def transcribe_audio(audio_file_path):
|
| 39 |
+
"""
|
| 40 |
+
Transcribes an audio file and generates an Excel file with segment details.
|
| 41 |
+
Note: This version does not handle files larger than 25MB by splitting them.
|
| 42 |
+
|
| 43 |
+
Args:
|
| 44 |
+
audio_file_path (str): The path to the input audio file.
|
| 45 |
+
|
| 46 |
+
Returns:
|
| 47 |
+
tuple: (str, str) - The full transcribed text and the path to the Excel file.
|
| 48 |
+
Returns error messages if transcription fails.
|
| 49 |
+
"""
|
| 50 |
+
if not audio_file_path:
|
| 51 |
+
return "Please upload an audio file.", None
|
| 52 |
+
|
| 53 |
+
# Check if the OpenAI API key is available
|
| 54 |
+
openai_api_key = os.getenv("OPENAI_API_KEY")
|
| 55 |
+
if not openai_api_key:
|
| 56 |
+
return "Error: OpenAI API Key not found. Please set the OPENAI_API_KEY environment variable.", None
|
| 57 |
+
|
| 58 |
+
try:
|
| 59 |
+
client = OpenAI(api_key=openai_api_key)
|
| 60 |
+
all_segments_data = [] # To store data for the Excel spreadsheet
|
| 61 |
+
full_text_transcription_parts = [] # To store text parts for the combined transcription
|
| 62 |
+
|
| 63 |
+
# Get file size for informational purposes (no chunking based on it)
|
| 64 |
+
file_size = os.path.getsize(audio_file_path)
|
| 65 |
+
print(f"Input file size: {file_size / (1024 * 1024):.2f} MB")
|
| 66 |
+
if file_size > MAX_FILE_SIZE_BYTES:
|
| 67 |
+
print(f"Warning: File size ({file_size / (1024 * 1024):.2f} MB) exceeds OpenAI's 25MB limit. Transcription might fail.")
|
| 68 |
+
print("Consider manually splitting very large files or reducing audio quality before upload.")
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
print(f"Transcribing: {audio_file_path} using OpenAI Whisper...")
|
| 72 |
+
transcription = transcribe_verbose(client, audio_file_path) # Direct transcription
|
| 73 |
+
|
| 74 |
+
# The OpenAI client returns an object where segments are accessed as attributes, not dictionary keys.
|
| 75 |
+
if hasattr(transcription, 'segments') and transcription.segments:
|
| 76 |
+
for segment in transcription.segments:
|
| 77 |
+
all_segments_data.append({
|
| 78 |
+
"Start Time": format_time(segment.start), # Access as attribute
|
| 79 |
+
"End Time": format_time(segment.end), # Access as attribute
|
| 80 |
+
"Speaker": "Unknown", # Placeholder for speaker, requires diarization
|
| 81 |
+
"Speech": segment.text.strip() # Access as attribute
|
| 82 |
+
})
|
| 83 |
+
# Collect all segment texts for the full transcription
|
| 84 |
+
full_text_transcription_parts.append(segment.text.strip())
|
| 85 |
+
else:
|
| 86 |
+
print("No segments found in transcription. Falling back to full text.")
|
| 87 |
+
# Fallback to full text if no segments are returned or segments attribute is missing
|
| 88 |
+
if hasattr(transcription, 'text'):
|
| 89 |
+
full_text_transcription_parts.append(transcription.text.strip())
|
| 90 |
+
else:
|
| 91 |
+
full_text_transcription_parts.append("No transcription text available.")
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
print("Transcription successful. Generating Excel.")
|
| 95 |
+
|
| 96 |
+
# Create Pandas DataFrame and export to Excel
|
| 97 |
+
df = pd.DataFrame(all_segments_data)
|
| 98 |
+
|
| 99 |
+
# Create a temporary Excel file
|
| 100 |
+
with tempfile.NamedTemporaryFile(suffix=".xlsx", delete=False) as temp_excel_file:
|
| 101 |
+
excel_path = temp_excel_file.name
|
| 102 |
+
df.to_excel(excel_path, index=False) # index=False prevents writing DataFrame index to Excel
|
| 103 |
+
|
| 104 |
+
# Join all segment texts into one continuous string for the main display
|
| 105 |
+
return " ".join(full_text_transcription_parts), excel_path
|
| 106 |
+
|
| 107 |
+
except Exception as e:
|
| 108 |
+
print(f"An error occurred during transcription: {e}")
|
| 109 |
+
return f"Error during transcription: {e}", None
|
| 110 |
+
|
| 111 |
+
# Create the Gradio interface
|
| 112 |
+
# gr.Audio for input, gr.Textbox for text output, gr.File for Excel download.
|
| 113 |
+
iface = gr.Interface(
|
| 114 |
+
fn=transcribe_audio,
|
| 115 |
+
inputs=gr.Audio(type="filepath", label="Upload Audio File (MP3, WAV, FLAC, etc.)"),
|
| 116 |
+
outputs=[
|
| 117 |
+
gr.Textbox(label="Full Transcription"),
|
| 118 |
+
gr.File(label="Download Transcription as Excel (.xlsx)")
|
| 119 |
+
],
|
| 120 |
+
title="Audio File Transcriber (OpenAI Whisper)",
|
| 121 |
+
description="Upload an audio file and get a text transcription using OpenAI's powerful Whisper model. The transcription will be displayed as one continuous text, and also available for download as an Excel spreadsheet with timestamps and speech segments. Note: Speaker identification is not included and would require a separate diarization model. Files larger than 25MB may fail due to OpenAI API limits.",
|
| 122 |
+
allow_flagging="never" # Disables the "Flag" button for this app
|
| 123 |
+
)
|
| 124 |
+
|
| 125 |
+
# Launch the Gradio application
|
| 126 |
+
if __name__ == "__main__":
|
| 127 |
+
print("Launching Gradio application...")
|
| 128 |
+
iface.launch()
|