app.py

# Gradio for Multi ASR
import gradio as gr
import torch
import soundfile as sf
from transformers import pipeline

# Device setup
device = "cuda:0" if torch.cuda.is_available() else "cpu"

# Pipeline cache
pipelines = {}

def get_pipeline(language):
    """Load and cache model pipelines based on selected language."""
    if language in pipelines:
        return pipelines[language]

    if language == "English":
        model_name = "openai/whisper-small"
    elif language == "Hindi":
        model_name = "vasista22/whisper-hindi-small"
    elif language == "Tamil":
        model_name = "vasista22/whisper-tamil-small"
    elif language == "Malayalam":
        model_name = "vrclc/W2V2-BERT-withLM-Malayalam-Studio"
    else:
        raise ValueError("Unsupported language")

    print(f"[INFO] Loading model for {language}: {model_name}")
    pipelines[language] = pipeline(
        "automatic-speech-recognition",
        model=model_name,
        device=device,
        chunk_length_s=10
    )
    return pipelines[language]

# Transcription code with error debugging
def transcribe(audio, language):
    """Transcribes speech from an audio file based on selected language."""
    try:
        if audio is None:
            return "Please record or upload an audio file."

        print(f"[DEBUG] Language: {language}")
        print(f"[DEBUG] Received audio: {audio}")

        audio_path = audio if isinstance(audio, str) else audio.get("name", None)
        if audio_path is None:
            return "Could not read audio file."

        audio_data, sample_rate = sf.read(audio_path)
        print(f"[DEBUG] Sample rate: {sample_rate}, shape: {audio_data.shape}")

        pipe = get_pipeline(language)

        # Prepare input format for transformers pipeline
        input_data = {"array": audio_data, "sampling_rate": sample_rate}

        result = pipe(input_data)["text"]
        print(f"[DEBUG] Transcription: {result}")
        return result

    except Exception as e:
        import traceback
        print("[ERROR] Exception during transcription:")
        traceback.print_exc()
        return f"Error: {str(e)}"

iface = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(sources=["microphone", "upload"], type="filepath", label="Record or Upload Audio"),
        gr.Dropdown(choices=["English", "Hindi", "Tamil", "Malayalam"], label="Select Language")
    ],
    outputs=gr.Textbox(label="Transcribed Text"),
    title="Multilingual Speech Recognition",
    description="Select a language and provide speech input to get transcription."
)

iface.launch()