Fastspeech2_HS_Flask_API / flask_app.py
k-m-irfan's picture
start.sh update
d994a8a
from flask import Flask, render_template, request, send_file, jsonify
import requests
import json
import ssl
import logging
import sys
import os
import base64
import io
#replace the path with your hifigan path to import Generator from models.py
sys.path.append("hifigan")
import torch
from espnet2.bin.tts_inference import Text2Speech
from models import Generator
from scipy.io.wavfile import write
from meldataset import MAX_WAV_VALUE
from env import AttrDict
import json
import yaml
from text_preprocess_for_inference import TTSDurAlignPreprocessor
# import time
logging.basicConfig(filename='access.log', level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
SAMPLING_RATE = 22050
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
preprocessor = TTSDurAlignPreprocessor()
app = Flask(__name__)
def load_hifigan_vocoder(language, gender, device):
# Load HiFi-GAN vocoder configuration file and generator model for the specified language and gender
vocoder_config = f"vocoder/{gender}/aryan/hifigan/config.json"
vocoder_generator = f"vocoder/{gender}/aryan/hifigan/generator"
# Read the contents of the vocoder configuration file
with open(vocoder_config, 'r') as f:
data = f.read()
json_config = json.loads(data)
h = AttrDict(json_config)
torch.manual_seed(h.seed)
# Move the generator model to the specified device (CPU or GPU)
device = torch.device(device)
generator = Generator(h).to(device)
state_dict_g = torch.load(vocoder_generator, device)
generator.load_state_dict(state_dict_g['generator'])
generator.eval()
generator.remove_weight_norm()
# Return the loaded and prepared HiFi-GAN generator model
return generator
def load_fastspeech2_model(language, gender, device):
models_directory = "./models"
# updating the config.yaml fiel based on language and gender
with open(f"{models_directory}/{language}/{gender}/model/config.yaml", "r") as file:
config = yaml.safe_load(file)
# current_working_directory = os.getcwd()
feat="model/feats_stats.npz"
pitch="model/pitch_stats.npz"
energy="model/energy_stats.npz"
feat_path=os.path.join(models_directory,language,gender,feat)
pitch_path=os.path.join(models_directory,language,gender,pitch)
energy_path=os.path.join(models_directory,language,gender,energy)
config["normalize_conf"]["stats_file"] = feat_path
config["pitch_normalize_conf"]["stats_file"] = pitch_path
config["energy_normalize_conf"]["stats_file"] = energy_path
with open(f"{models_directory}/{language}/{gender}/model/config.yaml", "w") as file:
yaml.dump(config, file)
tts_model = f"{models_directory}/{language}/{gender}/model/model.pth"
tts_config = f"{models_directory}/{language}/{gender}/model/config.yaml"
return Text2Speech(train_config=tts_config, model_file=tts_model, device=device)
def text_synthesis(language, gender, sample_text, vocoder, MAX_WAV_VALUE, device, alpha=1):
# Perform Text-to-Speech synthesis
with torch.no_grad():
# Load the FastSpeech2 model for the specified language and gender
model = load_fastspeech2_model(language, gender, device)
# Generate mel-spectrograms from the input text using the FastSpeech2 model
out = model(sample_text, decode_conf={"alpha": alpha})
print("TTS Done")
x = out["feat_gen_denorm"].T.unsqueeze(0) * 2.3262
x = x.to(device)
# Use the HiFi-GAN vocoder to convert mel-spectrograms to raw audio waveforms
y_g_hat = vocoder(x)
audio = y_g_hat.squeeze()
audio = audio * MAX_WAV_VALUE
audio = audio.cpu().numpy().astype('int16')
# Return the synthesized audio
return audio
def setup_app():
genders = ['male','female']
# to make dummy calls in all languages available
languages = {'hindi': "नमस्ते",'malayalam': "ഹലോ",'manipuri': "হ্যালো",'marathi': "हॅलो",'kannada': "ಹಲೋ",'bodo': "हॅलो",'english': "Hello",'assamese': "হ্যালো",'tamil': "ஹலோ",'odia': "ହେଲୋ",'rajasthani': "हॅलो",'telugu': "హలో",'bengali': "হ্যালো",'gujarati': "હલો","punjabi":"ਸਤ ਸ੍ਰੀ ਅਕਾਲ","urdu":"ہیلو"}
vocoders = {}
for gender in genders:
vocoders[gender]={}
for language,text in languages.items():
# Load the HiFi-GAN vocoder with dynamic language and gender
vocoder = load_hifigan_vocoder(language, gender, device)
vocoders[gender][language] = vocoder
# dummy calls
print(f"making dummy call for {language} - {gender}")
try:
out = text_synthesis(language, gender, text, vocoder, MAX_WAV_VALUE, device)
except:
message = f"cannot make dummy call for {gender} - {language} <==================="
print(message.upper())
print("Server Started...")
return vocoders
vocoders = setup_app()
@app.route('/', methods=['GET'])
def main():
return "IITM_TTS_V2"
@app.route('/tts', methods=['GET', 'POST'], strict_slashes=False)
def tts():
try:
json_data = request.get_json()
text = json_data["input"]
if not isinstance(text,str):
input_type = type(text)
ret = jsonify(status='failure', reason=f"Unsupported input type {input_type}. Input text should be in string format.")
gender = json_data["gender"]
language = json_data["lang"].lower()
alpha = json_data["alpha"]
# Preprocess the sample text
preprocessed_text, phrases = preprocessor.preprocess(text, language, gender)
preprocessed_text = " ".join(preprocessed_text)
vocoder = vocoders[gender][language]
out = text_synthesis(language, gender, preprocessed_text, vocoder, MAX_WAV_VALUE, device, alpha=alpha)
# output_file = f"{language}_{gender}_output.wav"
# write(output_file, SAMPLING_RATE, out)
# audio_wav_bytes = base64.b64encode(open(output_file, "rb").read())
# avoid saving file on disk
output_stream = io.BytesIO()
write(output_stream, SAMPLING_RATE, out)
audio_wav_bytes = base64.b64encode(output_stream.getvalue())
ret = jsonify(status="success",audio=audio_wav_bytes.decode('utf-8'))
except Exception as err:
ret = jsonify(status="failure", reason=str(err))
return ret
if __name__ == '__main__':
app.run(host='0.0.0.0', port=5000, debug=True)