app.py

import gradio as gr
from transformers import pipeline, VitsModel, AutoTokenizer
import numpy as np
import torch
import scipy

# Load the pipeline for speech recognition and translation
pipe = pipeline(
    "automatic-speech-recognition",
    model="Baghdad99/saad-speech-recognition-hausa-audio-to-text",
    tokenizer="Baghdad99/saad-speech-recognition-hausa-audio-to-text"
)
translator = pipeline("text2text-generation", model="Baghdad99/saad-hausa-text-to-english-text")
model = VitsModel.from_pretrained("facebook/mms-tts-eng")
tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")

# Define the function to translate speech
def translate_speech(audio):
    # Separate the sample rate and the audio data
    sample_rate, audio_data = audio

    # Use the speech recognition pipeline to transcribe the audio
    output = pipe(audio_data)
    print(f"Output: {output}")  # Print the output to see what it contains

    # Check if the output contains 'text'
    if 'text' in output[0]:
        transcription = output[0]["text"]
    else:
        print("The output does not contain 'text'")
        return

    # Use the translation pipeline to translate the transcription
    translated_text = translator(transcription)
    print(f"Translated text: {translated_text}")  # Print the translated text to see what it contains

    # Use the VITS model to synthesize the translated text into speech
    inputs = tokenizer(translated_text[0]['translation_text'], return_tensors="pt")
    with torch.no_grad():
        output = model.generate(**inputs)

    # Save the synthesized speech to a WAV file
    scipy.io.wavfile.write("synthesized_speech.wav", rate=model.config.sampling_rate, data=output.float().numpy())

    print("Translated text:", translated_text[0]['translation_text'])
    print("Synthesized speech data shape:", output.shape)
    print("Sampling rate:", model.config.sampling_rate)

    return 16000, output.numpy()

# Define the Gradio interface
iface = gr.Interface(
    fn=translate_speech, 
    inputs=gr.inputs.Audio(source="microphone", type="numpy"), 
    outputs=gr.outputs.Audio(type="numpy"),
    title="Hausa to English Translation",
    description="Realtime demo for Hausa to English translation using speech recognition and text-to-speech synthesis."
)

iface.launch()