taco2/tts.py

@@ -5,8 +5,6 @@ import numpy as np
 import torch
 import pyaudio
 import klepto
-import argparse
-from pathlib import Path
 from .model import Tacotron2
 from glow import WaveGlow
 from .hparams import HParams
@@ -16,8 +14,7 @@ from .denoiser import Denoiser
 from .audio_processing import griffin_lim, postprocess_audio
 
 TTS_SAMPLE_RATE = 22050
-OUTPUT_SAMPLE_RATE = 22050
-# OUTPUT_SAMPLE_RATE = 16000
+OUTPUT_SAMPLE_RATE = 16000
 
 # config from
 # https://github.com/NVIDIA/waveglow/blob/master/config.json
@@ -119,7 +116,7 @@ class TTSModel(object):
         audio = audio.cpu().numpy()
 
         return postprocess_audio(
-            audio, tempo=0.6, src_rate=TTS_SAMPLE_RATE, dst_rate=OUTPUT_SAMPLE_RATE
+            audio, src_rate=TTS_SAMPLE_RATE, dst_rate=OUTPUT_SAMPLE_RATE
         )
 
 
@@ -136,7 +133,10 @@ def player_gen():
     return play_device
 
 
-def repl(tts_model):
+def repl():
+    tts_model = TTSModel(
+        "/Users/malar/Work/tacotron2_r4_83000.pt", "/Users/malar/Work/waveglow_484000"
+    )
     player = player_gen()
 
     def loop():
@@ -148,26 +148,7 @@ def repl(tts_model):
 
 
 def main():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.ArgumentDefaultsHelpFormatter
-    )
-    parser.add_argument(
-        "-t",
-        "--tacotron2_path",
-        type=Path,
-        default="./tacotron.pt",
-        help="Path to a tacotron2 model",
-    )
-    parser.add_argument(
-        "-w",
-        "--waveglow_path",
-        type=Path,
-        default="./waveglow_256channels.pt",
-        help="Path to a waveglow model",
-    )
-    args = parser.parse_args()
-    tts_model = TTSModel(**vars(args))
-    interactive_loop = repl(tts_model)
+    interactive_loop = repl()
     while True:
         interactive_loop()