Update toy server for the recent updates

server/README.md

@@ -2,8 +2,8 @@
 Steps to run:
 1. Download one of the models given on the main page. Click [here](https://drive.google.com/drive/folders/1Q6BKeEkZyxSGsocK2p_mqgzLwlNvbHFJ?usp=sharing) for the lastest model.
 2. Checkout the corresponding commit history or use ```server``` branch if you like to use the latest model.
-2. Set the paths and the other options in the file ```server/conf.json```.
-3. Run the server ```python server/server.py -c server/conf.json```. (Requires Flask)
-4. Go to ```localhost:[given_port]``` and enjoy.
+3. Set the paths and the other options in the file ```server/conf.json```.
+4. Run the server ```python server/server.py -c server/conf.json```. (Requires Flask)
+5. Go to ```localhost:[given_port]``` and enjoy.
 
 For high quality results, please use the library versions shown in the ```requirements.txt``` file.

server/conf.json

@@ -1,6 +1,6 @@
 {
-    "model_path":"../models/May-22-2018_03_24PM-e6112f7",
-    "model_name":"checkpoint_272976.pth.tar",
+    "model_path":"/home/erogol/projects/runs/2579/keep/November-04-2018_06+19PM-TTS-master-_tmp-debug/",
+    "model_name":"best_model.pth.tar",
     "model_config":"config.json",
     "port": 5002,
     "use_cuda": true

server/server.py

@@ -1,7 +1,7 @@
 #!flask/bin/python
 import argparse
 from synthesizer import Synthesizer
-from TTS.utils.generic_utils import load_config
+from utils.generic_utils import load_config
 from flask import Flask, Response, request, render_template, send_file
 
 parser = argparse.ArgumentParser()

server/synthesizer.py

@@ -5,10 +5,10 @@ import torch
 import scipy
 import numpy as np
 import soundfile as sf
-from TTS.utils.text import text_to_sequence
-from TTS.utils.generic_utils import load_config
-from TTS.utils.audio import AudioProcessor
-from TTS.models.tacotron import Tacotron
+from utils.text import text_to_sequence
+from utils.generic_utils import load_config
+from utils.audio import AudioProcessor
+from models.tacotron import Tacotron
 from matplotlib import pylab as plt
 
 
@@ -22,19 +22,8 @@ class Synthesizer(object):
         config = load_config(model_config)
         self.config = config
         self.use_cuda = use_cuda
-        self.model = Tacotron(config.embedding_size, config.num_freq,
-                              config.num_mels, config.r)
-        self.ap = AudioProcessor(
-            config.sample_rate,
-            config.num_mels,
-            config.min_level_db,
-            config.frame_shift_ms,
-            config.frame_length_ms,
-            config.preemphasis,
-            config.ref_level_db,
-            config.num_freq,
-            config.power,
-            griffin_lim_iters=60)
+        self.ap = AudioProcessor(**config.audio)
+        self.model = Tacotron(config.embedding_size, self.ap.num_freq, self.ap.num_mels, config.r)
         # load model state
         if use_cuda:
             cp = torch.load(self.model_file)
@@ -48,9 +37,8 @@ class Synthesizer(object):
         self.model.eval()
 
     def save_wav(self, wav, path):
-        wav *= 32767 / max(1e-8, np.max(np.abs(wav)))
-        librosa.output.write_wav(path, wav.astype(np.int16),
-                                 self.config.sample_rate)
+        # wav *= 32767 / max(1e-8, np.max(np.abs(wav)))
+        self.ap.save_wav(wav, path)
 
     def tts(self, text):
         text_cleaner = [self.config.text_cleaner]
@@ -70,7 +58,6 @@ class Synthesizer(object):
                 chars_var)
             linear_out = linear_out[0].data.cpu().numpy()
             wav = self.ap.inv_spectrogram(linear_out.T)
-            # wav = wav[:self.ap.find_endpoint(wav)]
             out = io.BytesIO()
             wavs.append(wav)
             wavs.append(np.zeros(10000))