app.py

import gradio as gr
import requests
import numpy as np
from pydub import AudioSegment
import io

# Define the Hugging Face Inference API URLs and headers
TTS_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/english_voice_tts"
TRANSLATION_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/saad-hausa-text-to-english-text"
headers = {"Authorization": "Bearer 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"}

# Load the Gradio model for speech recognition
asr_model = gr.load("models/Baghdad99/saad-speech-recognition-hausa-audio-to-text")

# Define the function to query the Hugging Face Inference API
def query(api_url, payload):
    response = requests.post(api_url, headers=headers, json=payload)
    return response.json()

# Define the function to translate speech
def translate_speech(audio_file):
    print(f"Type of audio: {type(audio_file)}, Value of audio: {audio_file}")  # Debug line

    # Use the ASR model to transcribe the audio
    transcription = asr_model.predict(audio_file.name)  # Change this line

    # Use the translation pipeline to translate the transcription
    translated_text = query(TRANSLATION_API_URL, {"inputs": transcription})

    # Use the TTS pipeline to synthesize the translated text
    response = requests.post(TTS_API_URL, headers=headers, json={"inputs": translated_text})
    audio_bytes = response.content

    # Convert the audio bytes to an audio segment
    audio_segment = AudioSegment.from_mp3(io.BytesIO(audio_bytes))  # Change this line

    # Convert the audio segment to a numpy array
    audio_data = np.array(audio_segment.get_array_of_samples())
    if audio_segment.channels == 2:
        audio_data = audio_data.reshape((-1, 2))

    return audio_data

# Define the Gradio interface
iface = gr.Interface(
    fn=translate_speech, 
    inputs=gr.inputs.File(type="file"),  # Change this line
    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()