Spaces:
Building
Building
| import gradio as gr | |
| import requests | |
| import Levenshtein | |
| import librosa | |
| import torch | |
| from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor | |
| def load_model(): | |
| MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-arabic" | |
| processor = Wav2Vec2Processor.from_pretrained(MODEL_ID) | |
| model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID) | |
| return processor, model | |
| processor, model = load_model() | |
| def transcribe_audio_hf(audio_path): | |
| """ | |
| Transcribes speech from an audio file using a pretrained Wav2Vec2 model. | |
| Args: | |
| audio_path (str): Path to the audio file. | |
| Returns: | |
| str: The transcription of the speech in the audio file. | |
| """ | |
| speech_array, sampling_rate = librosa.load(audio_path, sr=16000) | |
| input_values = processor(speech_array, sampling_rate=sampling_rate, return_tensors="pt", padding=True).input_values | |
| with torch.no_grad(): | |
| logits = model(input_values).logits | |
| predicted_ids = torch.argmax(logits, dim=-1) | |
| transcription = processor.batch_decode(predicted_ids)[0].strip() | |
| return transcription | |
| def levenshtein_similarity(transcription1, transcription2): | |
| """ | |
| Calculate the Levenshtein similarity between two transcriptions. | |
| Args: | |
| transcription1 (str): The first transcription. | |
| transcription2 (str): The second transcription. | |
| Returns: | |
| float: A normalized similarity score between 0 and 1, where 1 indicates identical transcriptions. | |
| """ | |
| distance = Levenshtein.distance(transcription1, transcription2) | |
| max_len = max(len(transcription1), len(transcription2)) | |
| return 1 - distance / max_len # Normalize to get similarity score | |
| def evaluate_audio_similarity(original_audio, user_audio): | |
| """ | |
| Compares the similarity between the transcription of an original audio file and a user's audio file. | |
| Args: | |
| original_audio (str): Path to the original audio file. | |
| user_audio (str): Path to the user's audio file. | |
| Returns: | |
| tuple: Transcriptions and Levenshtein similarity score. | |
| """ | |
| transcription_original = transcribe_audio_hf(original_audio) | |
| transcription_user = transcribe_audio_hf(user_audio) | |
| similarity_score_levenshtein = levenshtein_similarity(transcription_original, transcription_user) | |
| return transcription_original, transcription_user, similarity_score_levenshtein | |
| def perform_testing(original_audio, user_audio): | |
| if original_audio is not None and user_audio is not None: | |
| transcription_original, transcription_user, similarity_score = evaluate_audio_similarity(original_audio, user_audio) | |
| return ( | |
| f"**Original Transcription:** {transcription_original}", | |
| f"**User Transcription:** {transcription_user}", | |
| f"**Levenshtein Similarity Score:** {similarity_score:.2f}" | |
| ) | |
| # Gradio Interface | |
| with gr.Blocks() as app: | |
| gr.Markdown("# Audio Transcription and Similarity Checker") | |
| original_audio_upload = gr.Audio(label="Upload Original Audio", type="filepath") | |
| user_audio_upload = gr.Audio(label="Upload User Audio", type="filepath") | |
| upload_button = gr.Button("Perform Testing") | |
| output_original_transcription = gr.Markdown() | |
| output_user_transcription = gr.Markdown() | |
| output_similarity_score = gr.Markdown() | |
| upload_button.click( | |
| perform_testing, | |
| inputs=[original_audio_upload, user_audio_upload], | |
| outputs=[output_original_transcription, output_user_transcription, output_similarity_score] | |
| ) | |
| app.launch(share=True) | |