app.py

# app.py

import gradio as gr
from transformers import pipeline
import numpy as np
import librosa  # pip install librosa

# --- EDIT THIS: map display names to your HF Hub model IDs ---
language_models = {
    "Akan (Asanti Twi)":        "FarmerlineML/w2v-bert-2.0_twi_alpha_v1",
    "Ewe":                      "FarmerlineML/w2v-bert-2.0_ewe_2",
    "Kiswahili":                "FarmerlineML/w2v-bert-2.0_swahili_alpha",
    "Luganda":                  "FarmerlineML/w2v-bert-2.0_luganda",
    "Brazilian Portuguese":     "FarmerlineML/w2v-bert-2.0_brazilian_portugese_alpha",
    "Fante Kissi":              "misterkissi/w2v2-lg-xls-r-300m-fante", 
    "Runyankore Kissi":         "misterkissi/w2v2-lg-xls-r-300m-runyankore",
    # add more as needed
}

# Pre-load pipelines for each language on CPU (device=-1)
asr_pipelines = {
    lang: pipeline(
        task="automatic-speech-recognition",
        model=model_id,
        device=-1,            # force CPU usage
        chunk_length_s=30
    )
    for lang, model_id in language_models.items()
}


def transcribe(audio_path: str, language: str) -> str:
    """
    Load the audio via librosa (supports mp3, wav, flac, m4a, ogg, etc.),
    convert to mono, then run it through the chosen ASR pipeline.
    """
    if not audio_path:
        return "⚠️ Please upload or record an audio clip."

    # librosa.load returns a 1D np.ndarray (mono) and the sample rate
    speech, sr = librosa.load(audio_path, sr=None, mono=True)

    # Call the Hugging Face ASR pipeline
    result = asr_pipelines[language]({
        "sampling_rate": sr,
        "raw": speech
    })
    return result.get("text", "")


with gr.Blocks(title="🌐 Multilingual ASR Demo") as demo:
    gr.Markdown(
        """
        ## 🎙️ Multilingual Speech-to-Text   
        Upload an audio file (MP3, WAV, FLAC, M4A, OGG,…) or record via your microphone.  
        Then choose the language/model and hit **Transcribe**.
        """
    )

    with gr.Row():
        lang = gr.Dropdown(
            choices=list(language_models.keys()),
            value=list(language_models.keys())[0],
            label="Select Language / Model"
        )

    with gr.Row():
        audio = gr.Audio(
            sources=["upload", "microphone"],
            type="filepath",
            label="Upload or Record Audio"
        )

    btn = gr.Button("Transcribe")
    output = gr.Textbox(label="Transcription")

    btn.click(fn=transcribe, inputs=[audio, lang], outputs=output)

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