app.py

import gradio as gr
import librosa
import numpy as np
import soundfile as sf
from transformers import pipeline
import os
from huggingface_hub import login

# Retrieve token from environment variable
hugging_face_token = os.getenv("ASR_CEB_HUGGING_FACE_TOKEN")

# Login using the token
login(token=hugging_face_token)

asr_ceb = pipeline("automatic-speech-recognition", model = "sil-ai/wav2vec2-bloom-speech-ceb")
asr_whisper_large = pipeline("automatic-speech-recognition", model = "openai/whisper-large-v3")
asr_whisper_ceb = pipeline("automatic-speech-recognition",
                           model = "nlewins/whisper-small-translate-X-gen2-examples-quality-step4-1e-6")


def transcribe_speech(filepath):
    if filepath is None:
        gr.Warning("No audio found, please retry.")
        return ""
    _, sample_rate = librosa.load(filepath, sr = None)
    model_rate = asr_ceb.feature_extractor.sampling_rate
    if sample_rate != model_rate:
        filepath = resample_audio_for_processing(filepath, model_rate, sample_rate)

    output_ceb = asr_ceb(filepath)
    generate_kwargs = {
        # "language": "tagalog",#source language
        "task": "translate"
    }
    output_whisper_large_translate = asr_whisper_large(filepath, generate_kwargs = generate_kwargs)
    output_whisper_large = asr_whisper_large(filepath)
    output_whisper_ceb = asr_whisper_ceb(filepath)
    return (output_ceb["text"], output_whisper_large["text"], output_whisper_large_translate["text"],
            output_whisper_ceb["text"])


def resample_audio_for_processing(filepath, model_rate, sample_rate):
    print(f"Audio loaded with rate: {sample_rate} Hz while model requires rate: {model_rate} Hz")
    try:
        print("Resampling audio...")
        audio_data, sr = librosa.load(filepath, sr = None)  # Audio data will be a NumPy array
        # Ensure that audio_data is a NumPy array
        audio_data = np.array(audio_data)
        # Resample to 16kHz
        audio_resampled = librosa.resample(audio_data, orig_sr = sample_rate, target_sr = model_rate)
        # Save the resampled audio
        resampled_audio_path = 'resampled_audio.wav'
        sf.write(resampled_audio_path, audio_resampled, 16000)
        print("Audio resampled successfully.")
        return resampled_audio_path
    except Exception as e:
        print(f"Error resampling audio: {e}, processing with audio as is it !")
        return filepath


mic_transcribe = gr.Interface(
    fn = transcribe_speech,
    inputs = gr.Audio(sources = ["microphone"], type = "filepath"),
    outputs = [gr.Textbox(label = "Transcription CEB (sil-ai)"), gr.Textbox(label = "Transcription (openai)"),
               gr.Textbox(label = "Translation (openai)"),
               gr.Textbox(label = "Transcription (nlewins)")]
    , allow_flagging = "never")

file_transcribe = gr.Interface(
    fn = transcribe_speech,
    inputs = gr.Audio(sources = ["upload"], type = "filepath"),
    outputs = [gr.Textbox(label = "Transcription CEB (sil-ai)"), gr.Textbox(label = "Transcription (openai)"),
               gr.Textbox(label = "Translation (openai)"),
               gr.Textbox(label = "Translation (nlewins)")]
    , allow_flagging = "never",
)

demo = gr.TabbedInterface(
    [mic_transcribe, file_transcribe],
    ["Use your Microphone", "Upload Audio File"],
)

if __name__ == '__main__':
    demo.launch()