app.py

import gradio as gr
import requests
import soundfile as sf
import numpy as np
import tempfile
from pydub import AudioSegment

# Define the Hugging Face Inference API URLs and headers
ASR_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/saad-speech-recognition-hausa-audio-to-text"
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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"}

# 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):
    print(f"Type of audio: {type(audio)}, Value of audio: {audio}")  # Debug line

    # audio is a tuple (np.ndarray, int), we need to save it as a file
    sample_rate, audio_data = audio
    if isinstance(audio_data, np.ndarray) and len(audio_data.shape) == 1:  # if audio_data is 1D, reshape it to 2D
        audio_data = np.reshape(audio_data, (-1, 1))
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
        sf.write(f, audio_data, sample_rate)
        audio_file = f.name

    # Use the ASR pipeline to transcribe the audio
    with open(audio_file, "rb") as f:
        data = f.read()
    response = requests.post(ASR_API_URL, headers=headers, data=data)
    output = response.json()

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

    # 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_file(io.BytesIO(audio_bytes), format="wav")

    # 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.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()