Gemini_Audio_Transcription.md

Chunks audio files

Original code from Surachet.

import librosa
import soundfile as sf
from pathlib import Path
from pydub import AudioSegment

def convert_mp4_to_wav(input_path, output_path=None):
   """Convert MP4 video to WAV audio using librosa"""
   if output_path is None:
       output_path = input_path.rsplit('.', 1)[0] + '.wav'

   # Load audio and convert to wav
   y, sr = librosa.load(input_path)
   sf.write(output_path, y, sr)
   print(f"Converted {input_path} to {output_path}")

def split_audio(audio_path, output_folder, duration=300, overlap=30):
    """
    Split audio file into segments with specified duration and overlap interval.
    
    Parameters:
        audio_path (str): Path to input audio file
        output_folder (str): Directory to save segments
        duration (int): Duration of each segment in seconds (default: 300)
        overlap (int): Overlap interval between segments in seconds (default: 30)
    """
    # Create output directory using pathlib
    output_dir = Path(output_folder)
    output_dir.mkdir(exist_ok=True)

    print(f"Loading: {audio_path}")
    audio = AudioSegment.from_file(audio_path)

    # Convert to milliseconds
    duration_ms = duration * 1000
    overlap_ms = overlap * 1000
    
    # Calculate step size (duration - overlap)
    step_ms = duration_ms - overlap_ms
    
    # Process segments
    segments = []
    for i, start in enumerate(range(0, len(audio) - overlap_ms, step_ms)):
        end = start + duration_ms
        if end > len(audio):
            end = len(audio)

        segment = audio[start:end]
        path = output_dir / f"segment_{i+1:02d}.wav"
        segments.append((i, segment, path))

    # Export segments
    for i, segment, path in segments:
        segment.export(path, format="wav")
        print(f"Segment {i+1} saved: {path}")
    
    print("Audio splitting completed.")

Usage

input_file = "../../example_file.mp4"
convert_mp4_to_wav(input_file)  # convert the MP4 to WAV format
split_audio("example_file.wav", "segments")  # segment into chunks

Gemini Transcription

from pathlib import Path
import json
import google.generativeai as genai
import pandas as pd
from glob import glob
from tqdm.auto import tqdm

genai.configure(api_key="API_KEY")
model = genai.GenerativeModel(model_name="gemini-1.5-flash")
prompt = "Please transcribe the given recorded interview speech."


def transcribe_audio_folder(folder_path, prompt="Please transcribe the given recorded interview speech."):
    """
    Transcribe all WAV files in a folder using Google's Generative AI.
    
    Args:
        folder_path (str): Path to folder containing WAV files
        prompt (str): Prompt for the transcription (default: "Generate a transcript of the speech.")
    
    Returns:
        pd.DataFrame: DataFrame containing audio filenames and their transcriptions
    """
    # Ensure folder path is valid
    folder = Path(folder_path)
    if not folder.exists():
        raise ValueError(f"Folder not found: {folder_path}")
    
    # Get all WAV files in the folder
    audio_files = list(folder.glob("*.wav"))
    if not audio_files:
        raise ValueError(f"No WAV files found in {folder_path}")
    
    transcriptions = []
    
    # Process each audio file
    for audio_file in tqdm(audio_files, desc="Transcribing audio files"):
        try:            
            # Upload and process audio file
            audio_upload = genai.upload_file(str(audio_file))
            response = model.generate_content([prompt, audio_upload])
            transcription = response.text.replace("```json", "").replace("```", "").strip()
            transcriptions.append({
                "audio_file": audio_file.name,
                "text": transcription
            })
        except Exception as e:
            print(f"Error processing {audio_file.name}: {str(e)}")
            transcriptions.append({
                "audio_file": audio_file.name,
                "text": f"ERROR: {str(e)}"
            })

    # Convert to DataFrame
    return pd.DataFrame(transcriptions)

# Example usage:
df = transcribe_audio_folder("path/to/segments")
df.to_csv("transcriptions.csv", index=False)