Spaces:
Running
Running
File size: 7,789 Bytes
b3221cd 65b0e9f f12a02d ed05346 f12a02d 28c5b18 9f630a9 4c58b5c f12a02d 28c5b18 ed05346 c2f19a4 28c5b18 c2f19a4 9f630a9 c2f19a4 ed05346 28c5b18 ed05346 28c5b18 ed05346 28c5b18 65b0e9f 28c5b18 c2f19a4 4c58b5c 28c5b18 c2f19a4 28c5b18 c2f19a4 28c5b18 c2f19a4 28c5b18 c2f19a4 28c5b18 b4eae74 d0c5307 9f630a9 d0c5307 28c5b18 b4eae74 28c5b18 4c58b5c 45b101e 9f630a9 45b101e 9f630a9 45b101e 9f630a9 f12a02d b213656 f12a02d b213656 45b101e 0a2805f 45b101e 9f630a9 f12a02d 9f630a9 45b101e c2f19a4 28c5b18 9f630a9 45b101e 9f630a9 f12a02d 45b101e 9f630a9 45b101e 9f630a9 45b101e 9f630a9 b213656 0a2805f 45b101e 9f630a9 45b101e 9f630a9 b213656 9f630a9 28c5b18 c2f19a4 f12a02d 4c58b5c 45b101e f12a02d 45b101e 0a2805f 9f630a9 45b101e 9f630a9 45b101e 9f630a9 45b101e 4c58b5c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 |
import gradio as gr
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import os
import json
import gspread
from oauth2client.service_account import ServiceAccountCredentials
from datetime import datetime
from gtts import gTTS
import tempfile
import requests
# --- CONFIGURATION ---
MODEL_K2H_REPO = "ankitklakra/kurukh-to-hindi"
MODEL_H2K_REPO = "ankitklakra/hindi-to-kurukh"
SHEET_NAME = "Kurukh_Feedback_Log"
# --- LOAD MODELS ---
print("Loading Translation Models...")
tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
model_k2h = AutoModelForSeq2SeqLM.from_pretrained(MODEL_K2H_REPO)
model_h2k = AutoModelForSeq2SeqLM.from_pretrained(MODEL_H2K_REPO)
pipe_k2h = pipeline("text2text-generation", model=model_k2h, tokenizer=tokenizer)
pipe_h2k = pipeline("text2text-generation", model=model_h2k, tokenizer=tokenizer)
print("Loading Voice Model...")
asr_pipeline = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
# --- HELPER FUNCTIONS ---
def transliterate_to_hindi(text):
try:
url = "https://inputtools.google.com/request?text={}&itc=hi-t-i0-und&num=1"
response = requests.get(url.format(text))
result = response.json()
return result[1][0][1][0]
except:
return text
def save_to_sheet(original, translation, correction, direction):
try:
json_creds = os.getenv("GOOGLE_CREDENTIALS")
if not json_creds: return "โ ๏ธ Error: Credentials missing."
creds_dict = json.loads(json_creds)
scope = ["https://spreadsheets.google.com/feeds", "https://www.googleapis.com/auth/drive"]
creds = ServiceAccountCredentials.from_json_keyfile_dict(creds_dict, scope)
client = gspread.authorize(creds)
sheet = client.open(SHEET_NAME).sheet1
if not sheet.get_all_values():
sheet.append_row(["Timestamp", "Direction", "Original Text", "AI Translation", "User Correction"])
sheet.append_row([str(datetime.now()), direction, original, translation, correction])
return "โ
Saved to Google Sheets."
except Exception as e:
return f"โ Error: {str(e)}"
def speech_to_text(audio_path):
if audio_path is None: return ""
return asr_pipeline(audio_path)["text"]
def text_to_speech(text, language="hi"):
if not text: return None
try:
tts = gTTS(text=text, lang=language)
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
tts.save(temp_file.name)
return temp_file.name
except: return None
# --- MAIN TRANSLATION LOGIC ---
def process_translation(text, audio_input, direction, is_hinglish):
original_text = speech_to_text(audio_input) if audio_input else text
if not original_text: return "", "", None
if direction == "Hindi -> Kurukh" and is_hinglish:
original_text = transliterate_to_hindi(original_text)
target_pipeline = pipe_k2h if direction == "Kurukh -> Hindi" else pipe_h2k
try:
results = target_pipeline(
original_text,
max_length=128,
num_beams=5,
no_repeat_ngram_size=2,
repetition_penalty=2.0,
early_stopping=True
)
translated_text = results[0]['generated_text']
except Exception as e:
return str(e), "", None
audio_output = None
if direction == "Kurukh -> Hindi":
audio_output = text_to_speech(translated_text, "hi")
return original_text, translated_text, audio_output
# --- CUSTOM CSS ---
custom_css = """
<style>
@import url('https://fonts.googleapis.com/css2?family=Poppins:wght@300;400;600&display=swap');
body, button, input, select, textarea, .gradio-container { font-family: 'Poppins', sans-serif !important; }
.header-div { text-align: center; margin-bottom: 20px; }
.header-title { font-size: 2.5em; font-weight: 600; color: #2c3e50; margin: 0; }
.header-subtitle { font-size: 1.2em; color: #7f8c8d; font-weight: 300; margin-top: 5px; }
</style>
"""
# --- THE UI ---
with gr.Blocks(theme=gr.themes.Soft(), title="Kurukh AI Translator") as demo:
# INJECT CSS
gr.HTML(custom_css)
# HEADER
gr.HTML("""
<div class="header-div">
<h1 class="header-title">๐ฎ๐ณ AI Kurukh (Oraon) Translator</h1>
<p class="header-subtitle">Bridging Communities with Artificial Intelligence | Voice & Hinglish Supported</p>
</div>
""")
with gr.Tabs():
# --- TAB 1: TRANSLATOR ---
with gr.TabItem("๐ฃ๏ธ Translator"):
with gr.Accordion("โน๏ธ How to use (Click to expand)", open=False):
gr.Markdown("""
1. **Select Mode:** Kurukh -> Hindi OR Hindi -> Kurukh.
2. **Hinglish:** Check the box if you type Hindi in English (e.g., 'Tumhara').
3. **Voice:** Click the microphone to speak (Hindi only).
""")
with gr.Row():
# LEFT COLUMN
with gr.Column():
gr.Markdown("### ๐ฅ Input Source")
with gr.Row():
direction = gr.Radio(["Kurukh -> Hindi", "Hindi -> Kurukh"], label="Translation Mode", value="Kurukh -> Hindi")
is_hinglish = gr.Checkbox(label="๐ค Hinglish Typing (e.g. 'Tumhara')", value=False)
input_text = gr.Textbox(label="Enter Text", placeholder="Type sentences here...", lines=4)
input_audio = gr.Audio(sources=["microphone"], type="filepath", label="๐๏ธ Voice Input (Hindi Only)")
translate_btn = gr.Button("Translate ๐", variant="primary")
# RIGHT COLUMN
with gr.Column():
gr.Markdown("### ๐ค Translation Result")
output_text = gr.Textbox(label="Translation", lines=4, interactive=False, show_copy_button=True)
output_audio = gr.Audio(label="๐ Listen (Hindi Only)", interactive=False, autoplay=False)
translate_btn.click(
fn=process_translation,
inputs=[input_text, input_audio, direction, is_hinglish],
outputs=[input_text, output_text, output_audio]
)
# --- TAB 2: FEEDBACK ---
with gr.TabItem("๐ Improve the AI"):
gr.Markdown("### ๐ ๏ธ Help us improve accuracy")
gr.Markdown("Our AI is learning! If you spot a mistake, please submit the correct translation below.")
with gr.Column():
with gr.Row():
fb_direction = gr.Radio(["Kurukh -> Hindi", "Hindi -> Kurukh"], label="Direction", value="Kurukh -> Hindi")
with gr.Row():
fb_original = gr.Textbox(label="Original Text")
fb_ai_output = gr.Textbox(label="AI's Wrong Translation")
fb_user_correct = gr.Textbox(label="โ
Your Correct Translation", lines=2, placeholder="Type the correct translation here")
submit_btn = gr.Button("Submit Correction to Database", variant="secondary")
status_lbl = gr.Label(label="Status")
submit_btn.click(
fn=save_to_sheet,
inputs=[fb_original, fb_ai_output, fb_user_correct, fb_direction],
outputs=status_lbl
)
gr.Markdown("---")
gr.HTML("<center style='color: #888;'>Built with โค๏ธ for the Kurukh Community โข Powered by Google mT5 & OpenAI Whisper</center>")
demo.launch() |