{
"cells": [
{
"cell_type": "code",
"execution_count": 55,
"id": "23e98a8a-7128-4f35-ba1c-ff514ed462e0",
"metadata": {},
"outputs": [],
"source": [
"#Install Dependencies\n",
"#!pip3 install torch torchvision torchaudio\n",
"#!pip install transformers ipywidgets gradio --upgrade\n",
"#!pip install --upgrade gradio\n",
"#!pip install nltk\n",
"#!pip install jiwer\n",
"#!pip install sentencepiece\n",
"#!pip install sacremoses\n",
"#!pip install soundfile"
]
},
{
"cell_type": "code",
"execution_count": 56,
"id": "29275fa9-1b88-4e37-a278-7118bfca860a",
"metadata": {},
"outputs": [],
"source": [
"\n",
"##translation_pipeline = pipeline('translation_en_to_fr')\n",
"##Evaluation Metric = BLEU score\n",
"##Exp1\n",
"#model_name = \"Davlan/byt5-base-eng-yor-mt\"\n",
"##Exp2\n",
"#model_name = \"Davlan/m2m100_418M-eng-yor-mt\" \n",
"##Exp3\n",
"#model_name = \"Davlan/mbart50-large-eng-yor-mt\"\n",
"##Exp4\n",
"#model_name = \"Davlan/mt5_base_eng_yor_mt\"\n",
"##Exp5\n",
"#model_name = \"omoekan/opus-tatoeba-eng-yor\"\n",
"##Exp6\n",
"#model_name = \"masakhane/afrimt5_en_yor_news\"\n",
"##Exp7\n",
"#model_name = \"masakhane/afrimbart_en_yor_news\"\n",
"##Exp8\n",
"#model_name = \"masakhane/afribyt5_en_yor_news\"\n",
"##Exp9\n",
"#model_name = \"masakhane/byt5_en_yor_news\"\n",
"##Exp10\n",
"#model_name = \"masakhane/mt5_en_yor_news\"\n",
"#translation_pipeline = pipeline('translation_en_to_yo', model = model_name, max_length=50)"
]
},
{
"cell_type": "code",
"execution_count": 57,
"id": "1ea4a2eb-6cbf-497a-a080-2db3dd64be36",
"metadata": {},
"outputs": [],
"source": [
"#results = translation_pipeline('My Name is Ayo, I love books')\n",
"#results[0]['translation_text']"
]
},
{
"cell_type": "code",
"execution_count": 58,
"id": "f92487b5-158a-47ef-ab12-a361ea8d0a48",
"metadata": {},
"outputs": [],
"source": [
"#results = translation_pipeline('The wages of sin is death')\n",
"#results[0]['translation_text']"
]
},
{
"cell_type": "code",
"execution_count": 59,
"id": "69d64db9-b083-46ae-80ce-9616ba99183d",
"metadata": {},
"outputs": [],
"source": [
"from transformers import pipeline\n",
"import nltk\n",
"import jiwer\n",
"from nltk.translate.bleu_score import corpus_bleu\n",
"from transformers import VitsModel, AutoTokenizer\n",
"import torch\n",
"import soundfile as sf\n",
"\n",
"\n",
"WerScore = 0\n",
"bleuScore = 0\n",
"def translate_transformers(modelName, sourceLangText):\n",
" #results = translation_pipeline(input_text)\n",
" translation_pipeline = pipeline('translation_en_to_yo', model = modelName, max_length=500)\n",
" translated_text = translation_pipeline(sourceLangText) #translator(text)[0][\"translation_text\"]\n",
" translated_text_target = translated_text[0]['translation_text']\n",
" hypothesis_translations = \"My name is Joy, I love reading\"\n",
" \n",
" #TTS for the translated_text_target\n",
" #TTS Exp1\n",
" ttsModel = VitsModel.from_pretrained(\"facebook/mms-tts-yor\")\n",
" tokenizer = AutoTokenizer.from_pretrained(\"facebook/mms-tts-yor\")\n",
" ttsInputs = tokenizer(translated_text_target, return_tensors=\"pt\")\n",
" \n",
" with torch.no_grad():\n",
" ttsOutput = ttsModel(**ttsInputs).waveform\n",
" #onvert the tensor to a numpy array\n",
" ttsWaveform = ttsOutput.numpy()[0] \n",
" #Save the waveform to an audio file\n",
" #sf.write('output.wav', waveform, 22050)\n",
" sf.write('ttsOutput.wav', ttsWaveform, 16000)\n",
" \n",
" #Calculate WerScore\n",
" WerScore = jiwer.wer(translated_text_target, hypothesis_translations)\n",
" #bleuScore = corpus_bleu(translated_text_target,hypothesis_translations)\n",
" \n",
" return translated_text_target,WerScore,'ttsOutput.wav'"
]
},
{
"cell_type": "code",
"execution_count": 60,
"id": "5d9ed5a2-0d28-4078-923d-c8c27196292a",
"metadata": {},
"outputs": [],
"source": [
"#text1 = \"Oruko mi ni Ayo, mo feran iwe kika gan\"\n",
"#text2 = \"Agbaninímọ̀ràn kan lórí ọ̀ràn radiation and Clinical Oncologist, tórúkọ rẹ̀ ń jẹ́ Temitope Olatunji-Agunbiade ti kìlọ̀ fáwọn obìnrin pé kí wọ́n má ṣe lo oògùn máàjóyúndúró tàbí kí wọ́n lo oògùn máàjóyúndúró, ó sọ pé ìwádìí ti fi hàn pé lílò tí wọ́n ń lò ó ń mú kí ewu àrùn jẹjẹrẹ ọmú pọ̀ sí i.\"\n",
"\n",
"#with torch.no_grad():\n",
" #output = ttsModel(**inputs).waveform"
]
},
{
"cell_type": "code",
"execution_count": 61,
"id": "54138308-b423-4e7c-9469-2002bfeb7918",
"metadata": {},
"outputs": [],
"source": [
"#from IPython.display import Audio\n",
"#Audio(output, rate=ttsModel.config.sampling_rate)"
]
},
{
"cell_type": "code",
"execution_count": 62,
"id": "bbf259d6-922d-4f5c-9af1-cbd57158a814",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Running on local URL: http://127.0.0.1:7879\n",
"Running on public URL: https://ccee705195aed67b23.gradio.live\n",
"\n",
"This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from Terminal to deploy to Spaces (https://huggingface.co/spaces)\n"
]
},
{
"data": {
"text/html": [
""
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 62,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"#Gradio Function and Interface\n",
"import gradio as gr\n",
"from IPython.display import Audio\n",
"interface = gr.Interface(\n",
" fn=translate_transformers,\n",
" inputs=[\n",
" gr.Dropdown([\"Davlan/byt5-base-eng-yor-mt\", #Exp1\n",
" \"Davlan/m2m100_418M-eng-yor-mt\", #Exp2\n",
" \"Davlan/mbart50-large-eng-yor-mt\", #Exp3\n",
" \"Davlan/mt5_base_eng_yor_mt\", #Exp4\n",
" \"omoekan/opus-tatoeba-eng-yor\", #Exp5\n",
" \"masakhane/afrimt5_en_yor_news\", #Exp6\n",
" \"masakhane/afrimbart_en_yor_news\", #Exp7\n",
" \"masakhane/afribyt5_en_yor_news\", #Exp8\n",
" \"masakhane/byt5_en_yor_news\", #Exp9\n",
" \"masakhane/mt5_en_yor_news\", #Exp10\n",
" \"masakhane/mbart50_en_yor_news\", #Exp11\n",
" \"masakhane/m2m100_418M_en_yor_news\", #Exp12\n",
" \"masakhane/m2m100_418M_en_yor_rel_news\", #Exp13\n",
" \"masakhane/m2m100_418M_en_yor_rel_news_ft\", #Exp14\n",
" \"masakhane/m2m100_418M_en_yor_rel\", #Exp15\n",
" #\"facebook/nllb-200-distilled-600M\", #Exp16\n",
" #\"facebook/nllb-200-3.3B\", #Exp17\n",
" #\"facebook/nllb-200-1.3B\", #Exp18\n",
" #\"facebook/nllb-200-distilled-1.3B\", #Exp19\n",
" #\"keithhon/nllb-200-3.3B\" #Exp20\n",
" #\"CohereForAI/aya-101\" #Exp16\n",
" ], \n",
" label=\"Select Finetuned Eng2Yor Translation Model\"),\n",
" gr.Textbox(lines=2, placeholder=\"Enter English Text Here...\", label=\"English Text\") \n",
" ],\n",
" #outputs = \"text\",\n",
" #outputs=outputs=[\"text\", \"text\"],#\"text\"\n",
" #outputs= gr.Textbox(value=\"text\", label=\"Translated Text\"),\n",
" outputs=[\n",
" gr.Textbox(value=\"text\", label=\"Translated Yoruba Text\"),\n",
" #gr.Textbox(value=\"text\", label=translated_text_actual),\n",
" gr.Textbox(value=\"number\", label=\"WER(Word Error Rate) Score - The Lower the Better\"),\n",
" #gr.Textbox(value=\"number\", label=\"Bleu Score\")\n",
" gr.Audio(type=\"filepath\", label=\"Click to Generate Yoruba Text2Speech\")\n",
" ],\n",
" title=\"ASPMIR NEURAL MACHINE TRANSLATION(NMT) TESTBED FOR LOW RESOURCED AFRICAN LANGUAGES\",\n",
" description=\"{This Tool Allows Developers and Researchers to Carry Out Experiments on Low Resourced African Languages with State-of-the-Art NMT Finetuned Models.}\"\n",
")\n",
"\n",
"interface.launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c3baee0f-fd85-4209-9d54-14451abd372a",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.10"
}
},
"nbformat": 4,
"nbformat_minor": 5
}