Spaces:
Running
Running
| import streamlit as st | |
| import firebase_admin | |
| import datetime | |
| import gradio as gr | |
| import numpy as np | |
| import tempfile | |
| from firebase_admin import credentials | |
| from firebase_admin import firestore | |
| from transformers import pipeline | |
| from typing import Optional | |
| from TTS.utils.manage import ModelManager | |
| from TTS.utils.synthesizer import Synthesizer | |
| from gradio import inputs | |
| from gradio.inputs import Textbox | |
| from gradio import outputs | |
| #Persistence via Cloud Store | |
| def get_db_firestore(): | |
| cred = credentials.Certificate('test.json') | |
| firebase_admin.initialize_app(cred, {'projectId': u'clinical-nlp-b9117',}) | |
| db = firestore.client() | |
| return db | |
| db = get_db_firestore() | |
| asr = pipeline("automatic-speech-recognition", "facebook/wav2vec2-base-960h") | |
| #STT Models | |
| MODEL_NAMES = [ | |
| "en/ljspeech/tacotron2-DDC", | |
| "en/ljspeech/glow-tts", | |
| "en/ljspeech/speedy-speech-wn", | |
| "en/ljspeech/vits", | |
| #"en/sam/tacotron-DDC", | |
| #"fr/mai/tacotron2-DDC", | |
| #"de/thorsten/tacotron2-DCA", | |
| ] | |
| MODELS = {} | |
| manager = ModelManager() | |
| for MODEL_NAME in MODEL_NAMES: | |
| print(f"downloading {MODEL_NAME}") | |
| model_path, config_path, model_item = manager.download_model(f"tts_models/{MODEL_NAME}") | |
| vocoder_name: Optional[str] = model_item["default_vocoder"] | |
| vocoder_path = None | |
| vocoder_config_path = None | |
| if vocoder_name is not None: | |
| vocoder_path, vocoder_config_path, _ = manager.download_model(vocoder_name) | |
| synthesizer = Synthesizer( | |
| model_path, config_path, None, vocoder_path, vocoder_config_path, | |
| ) | |
| MODELS[MODEL_NAME] = synthesizer | |
| GEN_NAMES = [ | |
| "huggingface/EleutherAI/gpt-neo-2.7B", | |
| "huggingface/EleutherAI/gpt-j-6B", | |
| "huggingface/gpt2-large" | |
| ] | |
| #ASR | |
| def transcribe(audio): | |
| text = asr(audio)["text"] | |
| return text | |
| #Sentiment Classifier | |
| classifier = pipeline("text-classification") | |
| # GPT-J: Story Generation Pipeline | |
| story_gen = pipeline("text-generation", "pranavpsv/gpt2-genre-story-generator") | |
| #STT | |
| def speech_to_text(speech): | |
| text = asr(speech)["text"] | |
| return text | |
| #TTSentiment | |
| def text_to_sentiment(text): | |
| sentiment = classifier(text)[0]["label"] | |
| return sentiment | |
| #Save | |
| def upsert(text): | |
| date_time =str(datetime.datetime.today()) | |
| doc_ref = db.collection('Text2SpeechSentimentSave').document(date_time) | |
| doc_ref.set({u'firefield': 'Recognize Speech', u'first': 'https://huggingface.co/spaces/awacke1/TTS-STT-Blocks/', u'last': text, u'born': date_time,}) | |
| saved = select('TTS-STT', date_time) | |
| # check it here: https://console.firebase.google.com/u/0/project/clinical-nlp-b9117/firestore/data/~2FStreamlitSpaces | |
| return saved | |
| #OpenLast | |
| def select(collection, document): | |
| doc_ref = db.collection(collection).document(document) | |
| doc = doc_ref.get() | |
| docid = ("The id is: ", doc.id) | |
| contents = ("The contents are: ", doc.to_dict()) | |
| return contents | |
| #OpenAll | |
| def selectall(text): | |
| docs = db.collection('Text2SpeechSentimentSave').stream() | |
| doclist='' | |
| for doc in docs: | |
| r=(f'{doc.id} => {doc.to_dict()}') | |
| doclist += r | |
| return doclist | |
| #TTS | |
| def tts(text: str, model_name: str): | |
| print(text, model_name) | |
| synthesizer = MODELS.get(model_name, None) | |
| if synthesizer is None: | |
| raise NameError("model not found") | |
| wavs = synthesizer.tts(text) | |
| with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp: | |
| synthesizer.save_wav(wavs, fp) | |
| return fp.name | |
| #Blocks Rock It | |
| demo = gr.Blocks() | |
| with demo: | |
| #UI | |
| audio_file = gr.inputs.Audio(source="microphone", type="filepath") | |
| text = gr.Textbox() | |
| label = gr.Label() | |
| saved = gr.Textbox() | |
| savedAll = gr.Textbox() | |
| TTSchoice = gr.inputs.Radio( label="Pick a TTS Model", choices=MODEL_NAMES, ) | |
| audio = gr.Audio(label="Output", interactive=False) | |
| #Buttons | |
| b1 = gr.Button("Recognize Speech") | |
| b2 = gr.Button("Classify Sentiment") | |
| b3 = gr.Button("Save Speech to Text") | |
| b4 = gr.Button("Retrieve All") | |
| b5 = gr.Button("Read It Back Aloud") | |
| #Event Model Chains | |
| b1.click(speech_to_text, inputs=audio_file, outputs=text) | |
| b2.click(text_to_sentiment, inputs=text, outputs=label) | |
| b3.click(upsert, inputs=text, outputs=saved) | |
| b4.click(selectall, inputs=text, outputs=savedAll) | |
| b5.click(tts, inputs=[text,TTSchoice], outputs=audio) | |
| # Lets Do It | |
| demo.launch(share=True) | |
| title = "Story Generators" | |
| examples = [ | |
| ["At which point do we invent Love?"], | |
| ["Love is a capacity more than consciousness is universal."], | |
| ["See the grace of god in eachother."], | |
| ["Love is a capacity more than consciousness is universal."], | |
| ["Love is generativity when there is more energy than what they need for equilibrium."], | |
| ["Collections of people have agency and mass having agency at the mesoscopic level"], | |
| ["Having a deep human connection is an interface problem to solve."], | |
| ["Having a collective creates agency since we build trust in eachother."] | |
| ] |