Merge pull request #45 from thorstenMueller/master

Added script createljspeech.py for easy dataset creation

README.md

@@ -97,6 +97,11 @@ Contributions are accepted! We'd love the communities help in building a better
    ```
    python3 preprocess.py --dataset ljspeech
    ```
+   If recorded with mimic-recording-studio
+   ````
+   python3 preprocess.py --dataset mrs --mrs_dir=<path_to>/mimic-recording-studio/
+   ````
+
      * other datasets can be used, i.e. `--dataset blizzard` for Blizzard data
      * for the mailabs dataset, do `preprocess.py --help` for options. Also, note that mailabs uses sample_size of 16000
      * you may want to create your own preprocessing script that works for your dataset. You can follow examples from preprocess.py and ./datasets
@@ -186,6 +191,14 @@ Contributions are accepted! We'd love the communities help in building a better
   * Here is the expected loss curve when training on LJ Speech with the default hyperparameters:
     ![Loss curve](https://user-images.githubusercontent.com/1945356/36077599-c0513e4a-0f21-11e8-8525-07347847720c.png)
 
+  * If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command
+  ````
+  python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-recording-studio/
+  ````
+  This generates an tacotron/LJSpeech-1.1 folder under your user home.
+
+
+
 ## Other Implementations
   * By Alex Barron: https://github.com/barronalex/Tacotron
   * By Kyubyong Park: https://github.com/Kyubyong/tacotron

datasets/createljspeech.py

@@ -0,0 +1,47 @@
+# This script generates a folder structure for ljspeech-1.1 processing from mimic-recording-studio database
+# Written by Thorsten Mueller (MrThorstenM@gmx.net) on november 2019 without any warranty
+
+import argparse
+import sqlite3
+import os
+from shutil import copyfile
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--out_dir', default=os.path.expanduser('~/tacotron'))
+    parser.add_argument('--mrs_dir', required=True)
+    args = parser.parse_args()
+
+    dir_base_ljspeech = os.path.join(args.out_dir,"LJSpeech-1.1")
+    dir_base_ljspeech_wav = os.path.join(dir_base_ljspeech,"wavs")
+    dir_base_mrs = args.mrs_dir
+    os.makedirs(dir_base_ljspeech_wav, exist_ok=True)
+
+    conn = sqlite3.connect(os.path.join(dir_base_mrs,"backend","db","mimicstudio.db"))
+    c = conn.cursor()
+
+    # Get user id from sqlite to find recordings in directory structure
+    # TODO: Currently just works with one user
+    for row in c.execute('SELECT uuid FROM usermodel LIMIT 1;'):
+        uid = row[0]
+  
+    print("Found speaker user guid in sqlite: " + uid)
+
+    # Create new metadata.csv for ljspeech
+    metadata = open(os.path.join(dir_base_ljspeech,"metadata.csv"),mode="w", encoding="utf8") 
+
+    for row in c.execute('SELECT DISTINCT audio_id, prompt, lower(prompt) FROM audiomodel ORDER BY length(prompt)'):
+        audio_file_source = os.path.join(dir_base_mrs,"backend","audio_files", uid, row[0] + ".wav")
+        if os.path.isfile(audio_file_source):
+            metadata.write(row[0] + "|" + row[1] + "|" + row[2] + "\n")
+            audio_file_dest = os.path.join(dir_base_ljspeech_wav,row[0] + ".wav")
+            copyfile(audio_file_source,audio_file_dest)
+        else:
+            print("File " + audio_file_source + " no found. Skipping.")
+
+
+    metadata.close()
+    conn.close()
+
+if __name__ == '__main__':
+    main()

datasets/mrs.py

@@ -0,0 +1,97 @@
+from concurrent.futures import ProcessPoolExecutor
+from functools import partial
+import numpy as np
+import os
+from util import audio
+import sqlite3
+import sys
+
+
+def build_from_path(in_dir, out_dir, username, num_workers=1, tqdm=lambda x: x):
+  '''Preprocesses the recordings from mimic-recording-studio (based on sqlite db) into a given output directory.
+
+    Args:
+      in_dir: The root directory of mimic-recording-studio
+      out_dir: The directory to write the output into
+      num_workers: Optional number of worker processes to parallelize across
+      tqdm: You can optionally pass tqdm to get a nice progress bar
+
+    Returns:
+      A list of tuples describing the training examples. This should be written to train.txt
+  '''
+
+  # We use ProcessPoolExecutor to parallize across processes. This is just an optimization and you
+  # can omit it and just call _process_utterance on each input if you want.
+  executor = ProcessPoolExecutor(max_workers=num_workers)
+  futures = []
+  index = 1
+
+  # Query sqlite db of mimic-recording-studio
+  dbfile = os.path.join(in_dir,"backend","db","mimicstudio.db")
+  print("Reading data from mimic-recording-studio sqlite db file: " + dbfile)
+  conn = sqlite3.connect(dbfile)
+  c = conn.cursor()
+
+  uid = ''
+  sql_get_guid = "SELECT uuid FROM usermodel LIMIT 1;"
+  if username:
+    print("Query user guid for " + username + " in sqlite db")
+    sql_get_guid = "SELECT uuid FROM usermodel WHERE UPPER(user_name) = '" + username.upper() + "' LIMIT 1;"
+
+  for row in c.execute(sql_get_guid):
+    uid = row[0]
+  
+  if uid == '':
+    print("No userid could be found in sqlite db.")
+    sys.exit()
+  
+  print("Found speaker user guid in sqlite: " + uid)
+
+  wav_dir = os.path.join(in_dir,"backend","audio_files",uid)
+  print("Search for wav files in " + wav_dir)
+  
+  for row in c.execute('SELECT DISTINCT audio_id, lower(prompt) FROM audiomodel ORDER BY length(prompt)'):
+    wav_path = os.path.join(wav_dir, '%s.wav' % row[0])
+    if os.path.isfile(wav_path):
+      text = row[1]
+      futures.append(executor.submit(partial(_process_utterance, out_dir, index, wav_path, text)))
+      index += 1
+    else:
+      print("File " + wav_path + " no found. Skipping.")
+  return [future.result() for future in tqdm(futures)]
+
+
+def _process_utterance(out_dir, index, wav_path, text):
+  '''Preprocesses a single utterance audio/text pair.
+
+  This writes the mel and linear scale spectrograms to disk and returns a tuple to write
+  to the train.txt file.
+
+  Args:
+    out_dir: The directory to write the spectrograms into
+    index: The numeric index to use in the spectrogram filenames.
+    wav_path: Path to the audio file containing the speech input
+    text: The text spoken in the input audio file
+
+  Returns:
+    A (spectrogram_filename, mel_filename, n_frames, text) tuple to write to train.txt
+  '''
+
+  # Load the audio to a numpy array:
+  wav = audio.load_wav(wav_path)
+
+  # Compute the linear-scale spectrogram from the wav:
+  spectrogram = audio.spectrogram(wav).astype(np.float32)
+  n_frames = spectrogram.shape[1]
+
+  # Compute a mel-scale spectrogram from the wav:
+  mel_spectrogram = audio.melspectrogram(wav).astype(np.float32)
+
+  # Write the spectrograms to disk:
+  spectrogram_filename = 'mrs-spec-%05d.npy' % index
+  mel_filename = 'mrs-mel-%05d.npy' % index
+  np.save(os.path.join(out_dir, spectrogram_filename), spectrogram.T, allow_pickle=False)
+  np.save(os.path.join(out_dir, mel_filename), mel_spectrogram.T, allow_pickle=False)
+
+  # Return a tuple describing this training example:
+  return (spectrogram_filename, mel_filename, n_frames, text)

preprocess.py

@@ -3,7 +3,9 @@ import os
 from multiprocessing import cpu_count
 from tqdm import tqdm
 from datasets import amy, blizzard, ljspeech, kusal, mailabs
+from datasets import mrs
 from hparams import hparams, hparams_debug_string
+import sys
 
 
 def preprocess_blizzard(args):
@@ -23,6 +25,14 @@ def preprocess_ljspeech(args):
       in_dir, out_dir, args.num_workers, tqdm=tqdm)
   write_metadata(metadata, out_dir)
 
+def preprocess_mrs(args):
+  in_dir = args.mrs_dir
+  out_dir = os.path.join(args.base_dir, args.output)
+  username = args.mrs_username
+  os.makedirs(out_dir, exist_ok=True)
+  metadata = mrs.build_from_path(
+      in_dir, out_dir, username, args.num_workers, tqdm=tqdm)
+  write_metadata(metadata, out_dir)
 
 def preprocess_amy(args):
   in_dir = os.path.join(args.base_dir, 'amy')
@@ -77,9 +87,11 @@ def write_metadata(metadata, out_dir):
 def main():
   parser = argparse.ArgumentParser()
   parser.add_argument('--base_dir', default=os.path.expanduser('~/tacotron'))
+  parser.add_argument('--mrs_dir', required=False)
+  parser.add_argument('--mrs_username', required=False)
   parser.add_argument('--output', default='training')
   parser.add_argument(
-      '--dataset', required=True, choices=['amy', 'blizzard', 'ljspeech', 'kusal', 'mailabs']
+      '--dataset', required=True, choices=['amy', 'blizzard', 'ljspeech', 'kusal', 'mailabs','mrs']
   )
   parser.add_argument('--mailabs_books_dir',
                       help='absolute directory to the books for the mlailabs')
@@ -109,6 +121,8 @@ def main():
     preprocess_kusal(args)
   elif args.dataset == 'mailabs':
     preprocess_mailabs(args)
+  elif args.dataset == 'mrs':
+    preprocess_mrs(args)
 
 
 if __name__ == "__main__":