| from gtts import gTTS | |
| from transformers import pipeline | |
| import gradio as gr | |
| import uuid | |
| asr = pipeline('automatic-speech-recognition', "facebook/wav2vec2-conformer-rope-large-960h-ft") | |
| corrector = pipeline("text2text-generation", model="pszemraj/grammar-synthesis-small") | |
| transcribe = lambda audio: asr(audio)['text'].lower() | |
| def to_audio(s): | |
| audio_path = f'/tmp/{uuid.uuid4()}.mp3' | |
| tts = gTTS(s, tld='us') | |
| tts.save(audio_path) | |
| return audio_path | |
| def transcription(audio, history): | |
| if audio: | |
| message = transcribe(audio) | |
| history.append(( (audio, ) , message)) | |
| results = corrector(message) | |
| results = '\n'.join([t['generated_text'] for t in results]) | |
| history.append( (None, f'**[Grammar and examples]**\n {results}') ) | |
| return history | |
| def chat(message, history): | |
| audio_path = to_audio(message) | |
| history.append((message, (audio_path,))) | |
| results = corrector(message) | |
| results = '\n'.join([t['generated_text'] for t in results]) | |
| history.append( (None, f'**[Grammar and examples]**\n {results}') ) | |
| return None, history | |
| with gr.Blocks(theme=gr.themes.Soft()) as learning: | |
| gr.Markdown('# The main aim of this app is to help English learners to speak fluently.') | |
| chatbot = gr.Chatbot() | |
| with gr.Row(): | |
| message = gr.Textbox(label='Send your message to TTS') | |
| microphone = gr.Audio(label="Transcribe", source="microphone", type="filepath") | |
| microphone.change(transcription, [microphone, chatbot], [chatbot]) | |
| microphone.change(lambda:None, None, microphone) | |
| message.submit(chat, [message, chatbot], [message, chatbot]) | |
| learning.launch() |