Update app.py

app.py

@@ -1,56 +1,49 @@
-import soundfile as sf
-import torch
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+# coding=utf8
+from gpt_index import SimpleDirectoryReader, GPTListIndex, GPTSimpleVectorIndex, LLMPredictor, PromptHelper
+from langchain import OpenAI
 import gradio as gr
-#import sox
-import subprocess
+import sys
+import os
 
+os.environ["OPENAI_API_KEY"] = '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'
 
-def read_file_and_process(wav_file):
-    filename = wav_file.split('.')[0]
-    filename_16k = filename + "16k.wav"
-    resampler(wav_file, filename_16k)
-    speech, _ = sf.read(filename_16k)
-    inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True)
 
-    return inputs
+def construct_index(directory_path):
+    max_input_size = 4096
+    num_outputs = 512
+    max_chunk_overlap = 20
+    chunk_size_limit = 600
 
+    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
 
-def resampler(input_file_path, output_file_path):
-    command = (
-        f"ffmpeg -hide_banner -loglevel panic -i {input_file_path} -ar 16000 -ac 1 -bits_per_raw_sample 16 -vn "
-        f"{output_file_path}"
-    )
-    subprocess.call(command, shell=True)
+    llm_predictor = LLMPredictor(llm=OpenAI(temperature=0.7, model_name="text-davinci-003", max_tokens=num_outputs))
 
+    documents = SimpleDirectoryReader(directory_path).load_data()
 
-def parse_transcription(logits):
-    predicted_ids = torch.argmax(logits, dim=-1)
-    transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
-    return transcription
+    index = GPTSimpleVectorIndex(documents, llm_predictor=llm_predictor, prompt_helper=prompt_helper)
 
+    index.save_to_disk('index.json')
 
-def parse(wav_file):
-    input_values = read_file_and_process(wav_file)
-    with torch.no_grad():
-        logits = model(**input_values).logits
-    return parse_transcription(logits)
+    return index
 
 
-model_id = "Harveenchadha/vakyansh-wav2vec2-hindi-him-4200"
-processor = Wav2Vec2Processor.from_pretrained(model_id)
-model = Wav2Vec2ForCTC.from_pretrained(model_id)
+def chatbot(input_text):
+    index = GPTSimpleVectorIndex.load_from_disk('index.json')
+    response = index.query(input_text, response_mode="compact")
+    return response.response
 
-input_ = gr.Audio(source="microphone", type="filepath")
-txtbox = gr.Textbox(
-    label="Hindi text output:",
-    lines=5
-)
+description = """
 
-title = "Speech-to-Text (Hindi) using Vakyansh"
-description = "Upload a hindi audio clip, and let AI do the hard work of transcribing."
-article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2104.06678'>Large-Scale Self- and Semi-Supervised Learning for Speech Translation</a></p>"
+<center>Olá sou a Zoh, fui treinada para responder perguntas com base das informações do Hippo Supermercados. Pergunte qualquer coisa. Caso eu não saiba, estarei aprendendo.
+<img src="https://s3.amazonaws.com/enlizt-resources-prod/companies/10958750-6306-11ea-b31c-2b332181af51_256_avatar?nocache=1588599205314" width=200px></center>
+"""
 
-gr.Interface(parse, inputs=input_, outputs=txtbox, title=title, description=description, article=article,
-             streaming=True, interactive=True,
-             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False,share=True);
+iface = gr.Interface(fn=chatbot,
+                     inputs=gr.inputs.Textbox(lines=3, label='O quê gostaria de saber?') ,
+                     outputs=gr.inputs.Textbox(lines=3, label="Veja o que encontrei"),
+                     description=description,
+                     css=".gradio-container-3-23-0 {background-color: #5f0000} .gradio-container-3-23-0 .prose * {color: #ffffff}",
+                     title="CD2 IA")
+
+index = construct_index(".")
+iface.launch(share=True, show_error=True)